2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1989, 1993
5 * The Regents of the University of California. All rights reserved.
7 * This code is derived from software contributed to Berkeley by
8 * Rick Macklem at The University of Guelph.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 #include <sys/cdefs.h>
37 __FBSDID("$FreeBSD$");
39 #include <sys/capsicum.h>
40 #include <sys/extattr.h>
43 * Functions that perform the vfs operations required by the routines in
44 * nfsd_serv.c. It is hoped that this change will make the server more
48 #include <fs/nfs/nfsport.h>
49 #include <security/mac/mac_framework.h>
50 #include <sys/filio.h>
52 #include <sys/sysctl.h>
53 #include <nlm/nlm_prot.h>
56 FEATURE(nfsd, "NFSv4 server");
58 extern u_int32_t newnfs_true, newnfs_false, newnfs_xdrneg1;
59 extern int nfsrv_useacl;
60 extern int newnfs_numnfsd;
61 extern struct mount nfsv4root_mnt;
62 extern struct nfsrv_stablefirst nfsrv_stablefirst;
63 extern void (*nfsd_call_servertimer)(void);
64 extern SVCPOOL *nfsrvd_pool;
65 extern struct nfsv4lock nfsd_suspend_lock;
66 extern struct nfsclienthashhead *nfsclienthash;
67 extern struct nfslockhashhead *nfslockhash;
68 extern struct nfssessionhash *nfssessionhash;
69 extern int nfsrv_sessionhashsize;
70 extern struct nfsstatsv1 nfsstatsv1;
71 extern struct nfslayouthash *nfslayouthash;
72 extern int nfsrv_layouthashsize;
73 extern struct mtx nfsrv_dslock_mtx;
74 extern int nfs_pnfsiothreads;
75 extern struct nfsdontlisthead nfsrv_dontlisthead;
76 extern volatile int nfsrv_dontlistlen;
77 extern volatile int nfsrv_devidcnt;
78 extern int nfsrv_maxpnfsmirror;
79 struct vfsoptlist nfsv4root_opt, nfsv4root_newopt;
82 struct nfsrchash_bucket nfsrchash_table[NFSRVCACHE_HASHSIZE];
83 struct nfsrchash_bucket nfsrcahash_table[NFSRVCACHE_HASHSIZE];
84 struct mtx nfsrc_udpmtx;
85 struct mtx nfs_v4root_mutex;
86 struct mtx nfsrv_dontlistlock_mtx;
87 struct mtx nfsrv_recalllock_mtx;
88 struct nfsrvfh nfs_rootfh, nfs_pubfh;
89 int nfs_pubfhset = 0, nfs_rootfhset = 0;
90 struct proc *nfsd_master_proc = NULL;
91 int nfsd_debuglevel = 0;
92 static pid_t nfsd_master_pid = (pid_t)-1;
93 static char nfsd_master_comm[MAXCOMLEN + 1];
94 static struct timeval nfsd_master_start;
95 static uint32_t nfsv4_sysid = 0;
96 static fhandle_t zerofh;
98 static int nfssvc_srvcall(struct thread *, struct nfssvc_args *,
101 int nfsrv_enable_crossmntpt = 1;
102 static int nfs_commit_blks;
103 static int nfs_commit_miss;
104 extern int nfsrv_issuedelegs;
105 extern int nfsrv_dolocallocks;
106 extern int nfsd_enable_stringtouid;
107 extern struct nfsdevicehead nfsrv_devidhead;
109 static int nfsrv_createiovec(int, struct mbuf **, struct mbuf **,
111 static int nfsrv_createiovecw(int, struct mbuf *, char *, struct iovec **,
113 static void nfsrv_pnfscreate(struct vnode *, struct vattr *, struct ucred *,
115 static void nfsrv_pnfsremovesetup(struct vnode *, NFSPROC_T *, struct vnode **,
116 int *, char *, fhandle_t *);
117 static void nfsrv_pnfsremove(struct vnode **, int, char *, fhandle_t *,
119 static int nfsrv_proxyds(struct vnode *, off_t, int, struct ucred *,
120 struct thread *, int, struct mbuf **, char *, struct mbuf **,
121 struct nfsvattr *, struct acl *, off_t *, int, bool *);
122 static int nfsrv_setextattr(struct vnode *, struct nfsvattr *, NFSPROC_T *);
123 static int nfsrv_readdsrpc(fhandle_t *, off_t, int, struct ucred *,
124 NFSPROC_T *, struct nfsmount *, struct mbuf **, struct mbuf **);
125 static int nfsrv_writedsrpc(fhandle_t *, off_t, int, struct ucred *,
126 NFSPROC_T *, struct vnode *, struct nfsmount **, int, struct mbuf **,
128 static int nfsrv_allocatedsrpc(fhandle_t *, off_t, off_t, struct ucred *,
129 NFSPROC_T *, struct vnode *, struct nfsmount **, int, int *);
130 static int nfsrv_setacldsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
131 struct vnode *, struct nfsmount **, int, struct acl *, int *);
132 static int nfsrv_setattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
133 struct vnode *, struct nfsmount **, int, struct nfsvattr *, int *);
134 static int nfsrv_getattrdsrpc(fhandle_t *, struct ucred *, NFSPROC_T *,
135 struct vnode *, struct nfsmount *, struct nfsvattr *);
136 static int nfsrv_seekdsrpc(fhandle_t *, off_t *, int, bool *, struct ucred *,
137 NFSPROC_T *, struct nfsmount *);
138 static int nfsrv_putfhname(fhandle_t *, char *);
139 static int nfsrv_pnfslookupds(struct vnode *, struct vnode *,
140 struct pnfsdsfile *, struct vnode **, NFSPROC_T *);
141 static void nfsrv_pnfssetfh(struct vnode *, struct pnfsdsfile *, char *, char *,
142 struct vnode *, NFSPROC_T *);
143 static int nfsrv_dsremove(struct vnode *, char *, struct ucred *, NFSPROC_T *);
144 static int nfsrv_dssetacl(struct vnode *, struct acl *, struct ucred *,
146 static int nfsrv_pnfsstatfs(struct statfs *, struct mount *);
148 int nfs_pnfsio(task_fn_t *, void *);
150 SYSCTL_NODE(_vfs, OID_AUTO, nfsd, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
152 SYSCTL_INT(_vfs_nfsd, OID_AUTO, mirrormnt, CTLFLAG_RW,
153 &nfsrv_enable_crossmntpt, 0, "Enable nfsd to cross mount points");
154 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_blks, CTLFLAG_RW, &nfs_commit_blks,
156 SYSCTL_INT(_vfs_nfsd, OID_AUTO, commit_miss, CTLFLAG_RW, &nfs_commit_miss,
158 SYSCTL_INT(_vfs_nfsd, OID_AUTO, issue_delegations, CTLFLAG_RW,
159 &nfsrv_issuedelegs, 0, "Enable nfsd to issue delegations");
160 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_locallocks, CTLFLAG_RW,
161 &nfsrv_dolocallocks, 0, "Enable nfsd to acquire local locks on files");
162 SYSCTL_INT(_vfs_nfsd, OID_AUTO, debuglevel, CTLFLAG_RW, &nfsd_debuglevel,
163 0, "Debug level for NFS server");
164 SYSCTL_INT(_vfs_nfsd, OID_AUTO, enable_stringtouid, CTLFLAG_RW,
165 &nfsd_enable_stringtouid, 0, "Enable nfsd to accept numeric owner_names");
166 static int nfsrv_pnfsgetdsattr = 1;
167 SYSCTL_INT(_vfs_nfsd, OID_AUTO, pnfsgetdsattr, CTLFLAG_RW,
168 &nfsrv_pnfsgetdsattr, 0, "When set getattr gets DS attributes via RPC");
171 * nfsrv_dsdirsize can only be increased and only when the nfsd threads are
173 * The dsN subdirectories for the increased values must have been created
174 * on all DS servers before this increase is done.
176 u_int nfsrv_dsdirsize = 20;
178 sysctl_dsdirsize(SYSCTL_HANDLER_ARGS)
180 int error, newdsdirsize;
182 newdsdirsize = nfsrv_dsdirsize;
183 error = sysctl_handle_int(oidp, &newdsdirsize, 0, req);
184 if (error != 0 || req->newptr == NULL)
186 if (newdsdirsize <= nfsrv_dsdirsize || newdsdirsize > 10000 ||
189 nfsrv_dsdirsize = newdsdirsize;
192 SYSCTL_PROC(_vfs_nfsd, OID_AUTO, dsdirsize,
193 CTLTYPE_UINT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, sizeof(nfsrv_dsdirsize),
194 sysctl_dsdirsize, "IU", "Number of dsN subdirs on the DS servers");
196 #define MAX_REORDERED_RPC 16
197 #define NUM_HEURISTIC 1031
198 #define NHUSE_INIT 64
200 #define NHUSE_MAX 2048
202 static struct nfsheur {
203 struct vnode *nh_vp; /* vp to match (unreferenced pointer) */
204 off_t nh_nextoff; /* next offset for sequential detection */
205 int nh_use; /* use count for selection */
206 int nh_seqcount; /* heuristic */
207 } nfsheur[NUM_HEURISTIC];
211 * Heuristic to detect sequential operation.
213 static struct nfsheur *
214 nfsrv_sequential_heuristic(struct uio *uio, struct vnode *vp)
219 /* Locate best candidate. */
221 hi = ((int)(vm_offset_t)vp / sizeof(struct vnode)) % NUM_HEURISTIC;
224 if (nfsheur[hi].nh_vp == vp) {
228 if (nfsheur[hi].nh_use > 0)
229 --nfsheur[hi].nh_use;
230 hi = (hi + 1) % NUM_HEURISTIC;
231 if (nfsheur[hi].nh_use < nh->nh_use)
235 /* Initialize hint if this is a new file. */
236 if (nh->nh_vp != vp) {
238 nh->nh_nextoff = uio->uio_offset;
239 nh->nh_use = NHUSE_INIT;
240 if (uio->uio_offset == 0)
246 /* Calculate heuristic. */
247 if ((uio->uio_offset == 0 && nh->nh_seqcount > 0) ||
248 uio->uio_offset == nh->nh_nextoff) {
249 /* See comments in vfs_vnops.c:sequential_heuristic(). */
250 nh->nh_seqcount += howmany(uio->uio_resid, 16384);
251 if (nh->nh_seqcount > IO_SEQMAX)
252 nh->nh_seqcount = IO_SEQMAX;
253 } else if (qabs(uio->uio_offset - nh->nh_nextoff) <= MAX_REORDERED_RPC *
254 imax(vp->v_mount->mnt_stat.f_iosize, uio->uio_resid)) {
255 /* Probably a reordered RPC, leave seqcount alone. */
256 } else if (nh->nh_seqcount > 1) {
257 nh->nh_seqcount /= 2;
261 nh->nh_use += NHUSE_INC;
262 if (nh->nh_use > NHUSE_MAX)
263 nh->nh_use = NHUSE_MAX;
268 * Get attributes into nfsvattr structure.
271 nfsvno_getattr(struct vnode *vp, struct nfsvattr *nvap,
272 struct nfsrv_descript *nd, struct thread *p, int vpislocked,
273 nfsattrbit_t *attrbitp)
275 int error, gotattr, lockedit = 0;
278 if (vpislocked == 0) {
280 * When vpislocked == 0, the vnode is either exclusively
281 * locked by this thread or not locked by this thread.
282 * As such, shared lock it, if not exclusively locked.
284 if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
286 NFSVOPLOCK(vp, LK_SHARED | LK_RETRY);
291 * Acquire the Change, Size, TimeAccess, TimeModify and SpaceUsed
292 * attributes, as required.
293 * This needs to be done for regular files if:
294 * - non-NFSv4 RPCs or
295 * - when attrbitp == NULL or
296 * - an NFSv4 RPC with any of the above attributes in attrbitp.
297 * A return of 0 for nfsrv_proxyds() indicates that it has acquired
298 * these attributes. nfsrv_proxyds() will return an error if the
299 * server is not a pNFS one.
302 if (vp->v_type == VREG && nfsrv_devidcnt > 0 && (attrbitp == NULL ||
303 (nd->nd_flag & ND_NFSV4) == 0 ||
304 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_CHANGE) ||
305 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SIZE) ||
306 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEACCESS) ||
307 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_TIMEMODIFY) ||
308 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEUSED))) {
309 error = nfsrv_proxyds(vp, 0, 0, nd->nd_cred, p,
310 NFSPROC_GETATTR, NULL, NULL, NULL, &na, NULL, NULL, 0,
316 error = VOP_GETATTR(vp, &nvap->na_vattr, nd->nd_cred);
321 * If we got the Change, Size and Modify Time from the DS,
325 nvap->na_atime = na.na_atime;
326 nvap->na_mtime = na.na_mtime;
327 nvap->na_filerev = na.na_filerev;
328 nvap->na_size = na.na_size;
329 nvap->na_bytes = na.na_bytes;
331 NFSD_DEBUG(4, "nfsvno_getattr: gotattr=%d err=%d chg=%ju\n", gotattr,
332 error, (uintmax_t)na.na_filerev);
339 * Get a file handle for a vnode.
342 nfsvno_getfh(struct vnode *vp, fhandle_t *fhp, struct thread *p)
346 NFSBZERO((caddr_t)fhp, sizeof(fhandle_t));
347 fhp->fh_fsid = vp->v_mount->mnt_stat.f_fsid;
348 error = VOP_VPTOFH(vp, &fhp->fh_fid);
355 * Perform access checking for vnodes obtained from file handles that would
356 * refer to files already opened by a Unix client. You cannot just use
357 * vn_writechk() and VOP_ACCESSX() for two reasons.
358 * 1 - You must check for exported rdonly as well as MNT_RDONLY for the write
360 * 2 - The owner is to be given access irrespective of mode bits for some
361 * operations, so that processes that chmod after opening a file don't
365 nfsvno_accchk(struct vnode *vp, accmode_t accmode, struct ucred *cred,
366 struct nfsexstuff *exp, struct thread *p, int override, int vpislocked,
367 u_int32_t *supportedtypep)
370 int error = 0, getret = 0;
372 if (vpislocked == 0) {
373 if (NFSVOPLOCK(vp, LK_SHARED) != 0) {
378 if (accmode & VWRITE) {
379 /* Just vn_writechk() changed to check rdonly */
381 * Disallow write attempts on read-only file systems;
382 * unless the file is a socket or a block or character
383 * device resident on the file system.
385 if (NFSVNO_EXRDONLY(exp) ||
386 (vp->v_mount->mnt_flag & MNT_RDONLY)) {
387 switch (vp->v_type) {
397 * If there's shared text associated with
398 * the inode, try to free it up once. If
399 * we fail, we can't allow writing.
401 if (VOP_IS_TEXT(vp) && error == 0)
411 * Should the override still be applied when ACLs are enabled?
413 error = VOP_ACCESSX(vp, accmode, cred, p);
414 if (error != 0 && (accmode & (VDELETE | VDELETE_CHILD))) {
416 * Try again with VEXPLICIT_DENY, to see if the test for
417 * deletion is supported.
419 error = VOP_ACCESSX(vp, accmode | VEXPLICIT_DENY, cred, p);
421 if (vp->v_type == VDIR) {
422 accmode &= ~(VDELETE | VDELETE_CHILD);
424 error = VOP_ACCESSX(vp, accmode, cred, p);
425 } else if (supportedtypep != NULL) {
426 *supportedtypep &= ~NFSACCESS_DELETE;
432 * Allow certain operations for the owner (reads and writes
433 * on files that are already open).
435 if (override != NFSACCCHK_NOOVERRIDE &&
436 (error == EPERM || error == EACCES)) {
437 if (cred->cr_uid == 0 && (override & NFSACCCHK_ALLOWROOT))
439 else if (override & NFSACCCHK_ALLOWOWNER) {
440 getret = VOP_GETATTR(vp, &vattr, cred);
441 if (getret == 0 && cred->cr_uid == vattr.va_uid)
454 * Set attribute(s) vnop.
457 nfsvno_setattr(struct vnode *vp, struct nfsvattr *nvap, struct ucred *cred,
458 struct thread *p, struct nfsexstuff *exp)
460 u_quad_t savsize = 0;
464 * If this is an exported file system and a pNFS service is running,
465 * don't VOP_SETATTR() of size for the MDS file system.
469 if (vp->v_type == VREG && (vp->v_mount->mnt_flag & MNT_EXPORTED) != 0 &&
470 nfsrv_devidcnt != 0 && nvap->na_vattr.va_size != VNOVAL &&
471 nvap->na_vattr.va_size > 0) {
472 savsize = nvap->na_vattr.va_size;
473 nvap->na_vattr.va_size = VNOVAL;
474 if (nvap->na_vattr.va_uid != (uid_t)VNOVAL ||
475 nvap->na_vattr.va_gid != (gid_t)VNOVAL ||
476 nvap->na_vattr.va_mode != (mode_t)VNOVAL ||
477 nvap->na_vattr.va_atime.tv_sec != VNOVAL ||
478 nvap->na_vattr.va_mtime.tv_sec != VNOVAL)
484 error = VOP_SETATTR(vp, &nvap->na_vattr, cred);
486 nvap->na_vattr.va_size = savsize;
487 if (error == 0 && (nvap->na_vattr.va_uid != (uid_t)VNOVAL ||
488 nvap->na_vattr.va_gid != (gid_t)VNOVAL ||
489 nvap->na_vattr.va_size != VNOVAL ||
490 nvap->na_vattr.va_mode != (mode_t)VNOVAL ||
491 nvap->na_vattr.va_atime.tv_sec != VNOVAL ||
492 nvap->na_vattr.va_mtime.tv_sec != VNOVAL)) {
493 /* For a pNFS server, set the attributes on the DS file. */
494 error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETATTR,
495 NULL, NULL, NULL, nvap, NULL, NULL, 0, NULL);
504 * Set up nameidata for a lookup() call and do it.
507 nfsvno_namei(struct nfsrv_descript *nd, struct nameidata *ndp,
508 struct vnode *dp, int islocked, struct nfsexstuff *exp, struct thread *p,
509 struct vnode **retdirp)
511 struct componentname *cnp = &ndp->ni_cnd;
515 int lockleaf = (cnp->cn_flags & LOCKLEAF) != 0, linklen;
520 cnp->cn_nameptr = cnp->cn_pnbuf;
523 * Extract and set starting directory.
525 if (dp->v_type != VDIR) {
530 nfsvno_relpathbuf(ndp);
538 if (NFSVNO_EXRDONLY(exp))
539 cnp->cn_flags |= RDONLY;
540 ndp->ni_segflg = UIO_SYSSPACE;
542 if (nd->nd_flag & ND_PUBLOOKUP) {
544 if (cnp->cn_pnbuf[0] == '/') {
547 * Check for degenerate pathnames here, since lookup()
550 for (i = 1; i < ndp->ni_pathlen; i++)
551 if (cnp->cn_pnbuf[i] != '/')
553 if (i == ndp->ni_pathlen) {
554 error = NFSERR_ACCES;
560 } else if ((nfsrv_enable_crossmntpt == 0 && NFSVNO_EXPORTED(exp)) ||
561 (nd->nd_flag & ND_NFSV4) == 0) {
563 * Only cross mount points for NFSv4 when doing a
564 * mount while traversing the file system above
565 * the mount point, unless nfsrv_enable_crossmntpt is set.
567 cnp->cn_flags |= NOCROSSMOUNT;
571 * Initialize for scan, set ni_startdir and bump ref on dp again
572 * because lookup() will dereference ni_startdir.
576 ndp->ni_startdir = dp;
577 ndp->ni_rootdir = rootvnode;
578 ndp->ni_topdir = NULL;
581 cnp->cn_flags |= LOCKLEAF;
583 cnp->cn_nameptr = cnp->cn_pnbuf;
585 * Call lookup() to do the real work. If an error occurs,
586 * ndp->ni_vp and ni_dvp are left uninitialized or NULL and
587 * we do not have to dereference anything before returning.
588 * In either case ni_startdir will be dereferenced and NULLed
596 * Check for encountering a symbolic link. Trivial
597 * termination occurs if no symlink encountered.
599 if ((cnp->cn_flags & ISSYMLINK) == 0) {
600 if ((cnp->cn_flags & (SAVENAME | SAVESTART)) == 0)
601 nfsvno_relpathbuf(ndp);
602 if (ndp->ni_vp && !lockleaf)
603 NFSVOPUNLOCK(ndp->ni_vp);
610 if ((cnp->cn_flags & LOCKPARENT) && ndp->ni_pathlen == 1)
611 NFSVOPUNLOCK(ndp->ni_dvp);
612 if (!(nd->nd_flag & ND_PUBLOOKUP)) {
617 if (ndp->ni_loopcnt++ >= MAXSYMLINKS) {
621 if (ndp->ni_pathlen > 1)
622 cp = uma_zalloc(namei_zone, M_WAITOK);
626 aiov.iov_len = MAXPATHLEN;
627 auio.uio_iov = &aiov;
630 auio.uio_rw = UIO_READ;
631 auio.uio_segflg = UIO_SYSSPACE;
633 auio.uio_resid = MAXPATHLEN;
634 error = VOP_READLINK(ndp->ni_vp, &auio, cnp->cn_cred);
637 if (ndp->ni_pathlen > 1)
638 uma_zfree(namei_zone, cp);
644 linklen = MAXPATHLEN - auio.uio_resid;
649 if (linklen + ndp->ni_pathlen >= MAXPATHLEN) {
650 error = ENAMETOOLONG;
655 * Adjust or replace path
657 if (ndp->ni_pathlen > 1) {
658 NFSBCOPY(ndp->ni_next, cp + linklen, ndp->ni_pathlen);
659 uma_zfree(namei_zone, cnp->cn_pnbuf);
662 cnp->cn_pnbuf[linklen] = '\0';
663 ndp->ni_pathlen += linklen;
666 * Cleanup refs for next loop and check if root directory
667 * should replace current directory. Normally ni_dvp
668 * becomes the new base directory and is cleaned up when
669 * we loop. Explicitly null pointers after invalidation
670 * to clarify operation.
675 if (cnp->cn_pnbuf[0] == '/') {
677 ndp->ni_dvp = ndp->ni_rootdir;
680 ndp->ni_startdir = ndp->ni_dvp;
684 cnp->cn_flags &= ~LOCKLEAF;
688 nfsvno_relpathbuf(ndp);
691 ndp->ni_startdir = NULL;
692 } else if ((ndp->ni_cnd.cn_flags & (WANTPARENT|LOCKPARENT)) == 0) {
697 NFSEXITCODE2(error, nd);
702 * Set up a pathname buffer and return a pointer to it and, optionally
703 * set a hash pointer.
706 nfsvno_setpathbuf(struct nameidata *ndp, char **bufpp, u_long **hashpp)
708 struct componentname *cnp = &ndp->ni_cnd;
710 cnp->cn_flags |= (NOMACCHECK | HASBUF);
711 cnp->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
714 *bufpp = cnp->cn_pnbuf;
718 * Release the above path buffer, if not released by nfsvno_namei().
721 nfsvno_relpathbuf(struct nameidata *ndp)
724 if ((ndp->ni_cnd.cn_flags & HASBUF) == 0)
726 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
727 ndp->ni_cnd.cn_flags &= ~HASBUF;
731 * Readlink vnode op into an mbuf list.
734 nfsvno_readlink(struct vnode *vp, struct ucred *cred, struct thread *p,
735 struct mbuf **mpp, struct mbuf **mpendp, int *lenp)
738 struct uio io, *uiop = &io;
739 struct mbuf *mp, *mp3;
740 int len, tlen, error = 0;
742 len = NFS_MAXPATHLEN;
743 uiop->uio_iovcnt = nfsrv_createiovec(len, &mp3, &mp, &iv);
745 uiop->uio_offset = 0;
746 uiop->uio_resid = len;
747 uiop->uio_rw = UIO_READ;
748 uiop->uio_segflg = UIO_SYSSPACE;
750 error = VOP_READLINK(vp, uiop, cred);
757 if (uiop->uio_resid > 0) {
758 len -= uiop->uio_resid;
759 tlen = NFSM_RNDUP(len);
763 } else if (tlen != NFS_MAXPATHLEN || tlen != len)
764 mp = nfsrv_adj(mp3, NFS_MAXPATHLEN - tlen,
777 * Create an mbuf chain and an associated iovec that can be used to Read
778 * or Getextattr of data.
779 * Upon success, return pointers to the first and last mbufs in the chain
780 * plus the malloc'd iovec and its iovlen.
783 nfsrv_createiovec(int len, struct mbuf **mpp, struct mbuf **mpendp,
786 struct mbuf *m, *m2 = NULL, *m3;
793 * Generate the mbuf list with the uio_iov ref. to it.
800 siz = min(M_TRAILINGSPACE(m), left);
809 *ivp = iv = malloc(i * sizeof (struct iovec), M_TEMP, M_WAITOK);
815 panic("nfsvno_read iov");
816 siz = min(M_TRAILINGSPACE(m), left);
818 iv->iov_base = mtod(m, caddr_t) + m->m_len;
833 * Read vnode op call into mbuf list.
836 nfsvno_read(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
837 struct thread *p, struct mbuf **mpp, struct mbuf **mpendp)
841 int error = 0, len, tlen, ioflag = 0;
843 struct uio io, *uiop = &io;
847 * Attempt to read from a DS file. A return of ENOENT implies
848 * there is no DS file to read.
850 error = nfsrv_proxyds(vp, off, cnt, cred, p, NFSPROC_READDS, mpp,
851 NULL, mpendp, NULL, NULL, NULL, 0, NULL);
855 len = NFSM_RNDUP(cnt);
856 uiop->uio_iovcnt = nfsrv_createiovec(len, &m3, &m, &iv);
858 uiop->uio_offset = off;
859 uiop->uio_resid = len;
860 uiop->uio_rw = UIO_READ;
861 uiop->uio_segflg = UIO_SYSSPACE;
863 nh = nfsrv_sequential_heuristic(uiop, vp);
864 ioflag |= nh->nh_seqcount << IO_SEQSHIFT;
865 /* XXX KDM make this more systematic? */
866 nfsstatsv1.srvbytes[NFSV4OP_READ] += uiop->uio_resid;
867 error = VOP_READ(vp, uiop, IO_NODELOCKED | ioflag, cred);
874 nh->nh_nextoff = uiop->uio_offset;
875 tlen = len - uiop->uio_resid;
876 cnt = cnt < tlen ? cnt : tlen;
877 tlen = NFSM_RNDUP(cnt);
881 } else if (len != tlen || tlen != cnt)
882 m = nfsrv_adj(m3, len - tlen, tlen - cnt);
892 * Create the iovec for the mbuf chain passed in as an argument.
893 * The "cp" argument is where the data starts within the first mbuf in
894 * the chain. It returns the iovec and the iovcnt.
897 nfsrv_createiovecw(int retlen, struct mbuf *m, char *cp, struct iovec **ivpp,
905 * Loop through the mbuf chain, counting how many mbufs are a
906 * part of this write operation, so the iovec size is known.
911 i = mtod(mp, caddr_t) + mp->m_len - cp;
925 /* Now, create the iovec. */
927 *ivpp = ivp = malloc(cnt * sizeof (struct iovec), M_TEMP,
930 i = mtod(mp, caddr_t) + mp->m_len - cp;
934 panic("nfsvno_write");
945 cp = mtod(mp, caddr_t);
952 * Write vnode op from an mbuf list.
955 nfsvno_write(struct vnode *vp, off_t off, int retlen, int *stable,
956 struct mbuf *mp, char *cp, struct ucred *cred, struct thread *p)
959 int cnt, ioflags, error;
960 struct uio io, *uiop = &io;
964 * Attempt to write to a DS file. A return of ENOENT implies
965 * there is no DS file to write.
967 error = nfsrv_proxyds(vp, off, retlen, cred, p, NFSPROC_WRITEDS,
968 &mp, cp, NULL, NULL, NULL, NULL, 0, NULL);
969 if (error != ENOENT) {
970 *stable = NFSWRITE_FILESYNC;
975 if (*stable == NFSWRITE_UNSTABLE)
976 ioflags = IO_NODELOCKED;
978 ioflags = (IO_SYNC | IO_NODELOCKED);
979 error = nfsrv_createiovecw(retlen, mp, cp, &iv, &cnt);
983 uiop->uio_iovcnt = cnt;
984 uiop->uio_resid = retlen;
985 uiop->uio_rw = UIO_WRITE;
986 uiop->uio_segflg = UIO_SYSSPACE;
988 uiop->uio_offset = off;
989 nh = nfsrv_sequential_heuristic(uiop, vp);
990 ioflags |= nh->nh_seqcount << IO_SEQSHIFT;
991 /* XXX KDM make this more systematic? */
992 nfsstatsv1.srvbytes[NFSV4OP_WRITE] += uiop->uio_resid;
993 error = VOP_WRITE(vp, uiop, ioflags, cred);
995 nh->nh_nextoff = uiop->uio_offset;
1003 * Common code for creating a regular file (plus special files for V2).
1006 nfsvno_createsub(struct nfsrv_descript *nd, struct nameidata *ndp,
1007 struct vnode **vpp, struct nfsvattr *nvap, int *exclusive_flagp,
1008 int32_t *cverf, NFSDEV_T rdev, struct nfsexstuff *exp)
1012 struct thread *p = curthread;
1014 error = nd->nd_repstat;
1015 if (!error && ndp->ni_vp == NULL) {
1016 if (nvap->na_type == VREG || nvap->na_type == VSOCK) {
1017 vrele(ndp->ni_startdir);
1018 error = VOP_CREATE(ndp->ni_dvp,
1019 &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1020 /* For a pNFS server, create the data file on a DS. */
1021 if (error == 0 && nvap->na_type == VREG) {
1023 * Create a data file on a DS for a pNFS server.
1024 * This function just returns if not
1025 * running a pNFS DS or the creation fails.
1027 nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr,
1031 nfsvno_relpathbuf(ndp);
1033 if (*exclusive_flagp) {
1034 *exclusive_flagp = 0;
1035 NFSVNO_ATTRINIT(nvap);
1036 nvap->na_atime.tv_sec = cverf[0];
1037 nvap->na_atime.tv_nsec = cverf[1];
1038 error = VOP_SETATTR(ndp->ni_vp,
1039 &nvap->na_vattr, nd->nd_cred);
1043 error = NFSERR_NOTSUPP;
1048 * NFS V2 Only. nfsrvd_mknod() does this for V3.
1049 * (This implies, just get out on an error.)
1051 } else if (nvap->na_type == VCHR || nvap->na_type == VBLK ||
1052 nvap->na_type == VFIFO) {
1053 if (nvap->na_type == VCHR && rdev == 0xffffffff)
1054 nvap->na_type = VFIFO;
1055 if (nvap->na_type != VFIFO &&
1056 (error = priv_check_cred(nd->nd_cred, PRIV_VFS_MKNOD_DEV))) {
1057 vrele(ndp->ni_startdir);
1058 nfsvno_relpathbuf(ndp);
1062 nvap->na_rdev = rdev;
1063 error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
1064 &ndp->ni_cnd, &nvap->na_vattr);
1066 nfsvno_relpathbuf(ndp);
1067 vrele(ndp->ni_startdir);
1071 vrele(ndp->ni_startdir);
1072 nfsvno_relpathbuf(ndp);
1080 * Handle cases where error is already set and/or
1082 * 1 - clean up the lookup
1083 * 2 - iff !error and na_size set, truncate it
1085 vrele(ndp->ni_startdir);
1086 nfsvno_relpathbuf(ndp);
1088 if (ndp->ni_dvp == *vpp)
1092 if (!error && nvap->na_size != VNOVAL) {
1093 error = nfsvno_accchk(*vpp, VWRITE,
1094 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1095 NFSACCCHK_VPISLOCKED, NULL);
1097 tempsize = nvap->na_size;
1098 NFSVNO_ATTRINIT(nvap);
1099 nvap->na_size = tempsize;
1100 error = VOP_SETATTR(*vpp,
1101 &nvap->na_vattr, nd->nd_cred);
1114 * Do a mknod vnode op.
1117 nfsvno_mknod(struct nameidata *ndp, struct nfsvattr *nvap, struct ucred *cred,
1123 vtyp = nvap->na_type;
1125 * Iff doesn't exist, create it.
1128 vrele(ndp->ni_startdir);
1129 nfsvno_relpathbuf(ndp);
1135 if (vtyp != VCHR && vtyp != VBLK && vtyp != VSOCK && vtyp != VFIFO) {
1136 vrele(ndp->ni_startdir);
1137 nfsvno_relpathbuf(ndp);
1139 error = NFSERR_BADTYPE;
1142 if (vtyp == VSOCK) {
1143 vrele(ndp->ni_startdir);
1144 error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp,
1145 &ndp->ni_cnd, &nvap->na_vattr);
1147 nfsvno_relpathbuf(ndp);
1149 if (nvap->na_type != VFIFO &&
1150 (error = priv_check_cred(cred, PRIV_VFS_MKNOD_DEV))) {
1151 vrele(ndp->ni_startdir);
1152 nfsvno_relpathbuf(ndp);
1156 error = VOP_MKNOD(ndp->ni_dvp, &ndp->ni_vp,
1157 &ndp->ni_cnd, &nvap->na_vattr);
1159 nfsvno_relpathbuf(ndp);
1160 vrele(ndp->ni_startdir);
1162 * Since VOP_MKNOD returns the ni_vp, I can't
1163 * see any reason to do the lookup.
1176 nfsvno_mkdir(struct nameidata *ndp, struct nfsvattr *nvap, uid_t saved_uid,
1177 struct ucred *cred, struct thread *p, struct nfsexstuff *exp)
1181 if (ndp->ni_vp != NULL) {
1182 if (ndp->ni_dvp == ndp->ni_vp)
1187 nfsvno_relpathbuf(ndp);
1191 error = VOP_MKDIR(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
1194 nfsvno_relpathbuf(ndp);
1205 nfsvno_symlink(struct nameidata *ndp, struct nfsvattr *nvap, char *pathcp,
1206 int pathlen, int not_v2, uid_t saved_uid, struct ucred *cred, struct thread *p,
1207 struct nfsexstuff *exp)
1212 vrele(ndp->ni_startdir);
1213 nfsvno_relpathbuf(ndp);
1214 if (ndp->ni_dvp == ndp->ni_vp)
1223 error = VOP_SYMLINK(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd,
1224 &nvap->na_vattr, pathcp);
1226 vrele(ndp->ni_startdir);
1227 nfsvno_relpathbuf(ndp);
1229 * Although FreeBSD still had the lookup code in
1230 * it for 7/current, there doesn't seem to be any
1231 * point, since VOP_SYMLINK() returns the ni_vp.
1232 * Just vput it for v2.
1234 if (!not_v2 && !error)
1243 * Parse symbolic link arguments.
1244 * This function has an ugly side effect. It will malloc() an area for
1245 * the symlink and set iov_base to point to it, only if it succeeds.
1246 * So, if it returns with uiop->uio_iov->iov_base != NULL, that must
1250 nfsvno_getsymlink(struct nfsrv_descript *nd, struct nfsvattr *nvap,
1251 struct thread *p, char **pathcpp, int *lenp)
1254 char *pathcp = NULL;
1256 struct nfsv2_sattr *sp;
1260 if ((nd->nd_flag & ND_NFSV3) &&
1261 (error = nfsrv_sattr(nd, NULL, nvap, NULL, NULL, p)))
1263 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
1264 len = fxdr_unsigned(int, *tl);
1265 if (len > NFS_MAXPATHLEN || len <= 0) {
1269 pathcp = malloc(len + 1, M_TEMP, M_WAITOK);
1270 error = nfsrv_mtostr(nd, pathcp, len);
1273 if (nd->nd_flag & ND_NFSV2) {
1274 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
1275 nvap->na_mode = fxdr_unsigned(u_int16_t, sp->sa_mode);
1279 NFSEXITCODE2(0, nd);
1283 free(pathcp, M_TEMP);
1284 NFSEXITCODE2(error, nd);
1289 * Remove a non-directory object.
1292 nfsvno_removesub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1293 struct thread *p, struct nfsexstuff *exp)
1295 struct vnode *vp, *dsdvp[NFSDEV_MAXMIRRORS];
1296 int error = 0, mirrorcnt;
1297 char fname[PNFS_FILENAME_LEN + 1];
1302 if (vp->v_type == VDIR)
1303 error = NFSERR_ISDIR;
1305 error = nfsrv_checkremove(vp, 1, NULL, (nfsquad_t)((u_quad_t)0),
1308 nfsrv_pnfsremovesetup(vp, p, dsdvp, &mirrorcnt, fname, &fh);
1310 error = VOP_REMOVE(ndp->ni_dvp, vp, &ndp->ni_cnd);
1311 if (error == 0 && dsdvp[0] != NULL)
1312 nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p);
1313 if (ndp->ni_dvp == vp)
1318 if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1319 nfsvno_relpathbuf(ndp);
1325 * Remove a directory.
1328 nfsvno_rmdirsub(struct nameidata *ndp, int is_v4, struct ucred *cred,
1329 struct thread *p, struct nfsexstuff *exp)
1335 if (vp->v_type != VDIR) {
1340 * No rmdir "." please.
1342 if (ndp->ni_dvp == vp) {
1347 * The root of a mounted filesystem cannot be deleted.
1349 if (vp->v_vflag & VV_ROOT)
1353 error = VOP_RMDIR(ndp->ni_dvp, vp, &ndp->ni_cnd);
1354 if (ndp->ni_dvp == vp)
1359 if ((ndp->ni_cnd.cn_flags & SAVENAME) != 0)
1360 nfsvno_relpathbuf(ndp);
1369 nfsvno_rename(struct nameidata *fromndp, struct nameidata *tondp,
1370 u_int32_t ndstat, u_int32_t ndflag, struct ucred *cred, struct thread *p)
1372 struct vnode *fvp, *tvp, *tdvp, *dsdvp[NFSDEV_MAXMIRRORS];
1373 int error = 0, mirrorcnt;
1374 char fname[PNFS_FILENAME_LEN + 1];
1378 fvp = fromndp->ni_vp;
1380 vrele(fromndp->ni_dvp);
1385 tdvp = tondp->ni_dvp;
1388 if (fvp->v_type == VDIR && tvp->v_type != VDIR) {
1389 error = (ndflag & ND_NFSV2) ? EISDIR : EEXIST;
1391 } else if (fvp->v_type != VDIR && tvp->v_type == VDIR) {
1392 error = (ndflag & ND_NFSV2) ? ENOTDIR : EEXIST;
1395 if (tvp->v_type == VDIR && tvp->v_mountedhere) {
1396 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1401 * A rename to '.' or '..' results in a prematurely
1402 * unlocked vnode on FreeBSD5, so I'm just going to fail that
1405 if ((tondp->ni_cnd.cn_namelen == 1 &&
1406 tondp->ni_cnd.cn_nameptr[0] == '.') ||
1407 (tondp->ni_cnd.cn_namelen == 2 &&
1408 tondp->ni_cnd.cn_nameptr[0] == '.' &&
1409 tondp->ni_cnd.cn_nameptr[1] == '.')) {
1414 if (fvp->v_type == VDIR && fvp->v_mountedhere) {
1415 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1418 if (fvp->v_mount != tdvp->v_mount) {
1419 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EXDEV;
1423 error = (ndflag & ND_NFSV2) ? ENOTEMPTY : EINVAL;
1428 * If source and destination are the same, there is nothing to
1429 * do. Set error to -1 to indicate this.
1434 if (ndflag & ND_NFSV4) {
1435 if (NFSVOPLOCK(fvp, LK_EXCLUSIVE) == 0) {
1436 error = nfsrv_checkremove(fvp, 0, NULL,
1437 (nfsquad_t)((u_quad_t)0), p);
1442 error = nfsrv_checkremove(tvp, 1, NULL,
1443 (nfsquad_t)((u_quad_t)0), p);
1446 * For NFSv2 and NFSv3, try to get rid of the delegation, so
1447 * that the NFSv4 client won't be confused by the rename.
1448 * Since nfsd_recalldelegation() can only be called on an
1449 * unlocked vnode at this point and fvp is the file that will
1450 * still exist after the rename, just do fvp.
1452 nfsd_recalldelegation(fvp, p);
1454 if (error == 0 && tvp != NULL) {
1455 nfsrv_pnfsremovesetup(tvp, p, dsdvp, &mirrorcnt, fname, &fh);
1456 NFSD_DEBUG(4, "nfsvno_rename: pnfsremovesetup"
1457 " dsdvp=%p\n", dsdvp[0]);
1461 error = VOP_RENAME(fromndp->ni_dvp, fromndp->ni_vp,
1462 &fromndp->ni_cnd, tondp->ni_dvp, tondp->ni_vp,
1471 vrele(fromndp->ni_dvp);
1478 * If dsdvp[0] != NULL, it was set up by nfsrv_pnfsremovesetup() and
1479 * if the rename succeeded, the DS file for the tvp needs to be
1482 if (error == 0 && dsdvp[0] != NULL) {
1483 nfsrv_pnfsremove(dsdvp, mirrorcnt, fname, &fh, p);
1484 NFSD_DEBUG(4, "nfsvno_rename: pnfsremove\n");
1487 vrele(tondp->ni_startdir);
1488 nfsvno_relpathbuf(tondp);
1490 vrele(fromndp->ni_startdir);
1491 nfsvno_relpathbuf(fromndp);
1500 nfsvno_link(struct nameidata *ndp, struct vnode *vp, struct ucred *cred,
1501 struct thread *p, struct nfsexstuff *exp)
1511 if (vp->v_mount != xp->v_mount)
1515 NFSVOPLOCK(vp, LK_EXCLUSIVE | LK_RETRY);
1516 if (!VN_IS_DOOMED(vp))
1517 error = VOP_LINK(ndp->ni_dvp, vp, &ndp->ni_cnd);
1520 if (ndp->ni_dvp == vp)
1526 if (ndp->ni_dvp == ndp->ni_vp)
1533 nfsvno_relpathbuf(ndp);
1539 * Do the fsync() appropriate for the commit.
1542 nfsvno_fsync(struct vnode *vp, u_int64_t off, int cnt, struct ucred *cred,
1548 * RFC 1813 3.3.21: if count is 0, a flush from offset to the end of
1549 * file is done. At this time VOP_FSYNC does not accept offset and
1550 * byte count parameters so call VOP_FSYNC the whole file for now.
1551 * The same is true for NFSv4: RFC 3530 Sec. 14.2.3.
1552 * File systems that do not use the buffer cache (as indicated
1553 * by MNTK_USES_BCACHE not being set) must use VOP_FSYNC().
1555 if (cnt == 0 || cnt > MAX_COMMIT_COUNT ||
1556 (vp->v_mount->mnt_kern_flag & MNTK_USES_BCACHE) == 0) {
1558 * Give up and do the whole thing
1560 if (vp->v_object && vm_object_mightbedirty(vp->v_object)) {
1561 VM_OBJECT_WLOCK(vp->v_object);
1562 vm_object_page_clean(vp->v_object, 0, 0, OBJPC_SYNC);
1563 VM_OBJECT_WUNLOCK(vp->v_object);
1565 error = VOP_FSYNC(vp, MNT_WAIT, td);
1568 * Locate and synchronously write any buffers that fall
1569 * into the requested range. Note: we are assuming that
1570 * f_iosize is a power of 2.
1572 int iosize = vp->v_mount->mnt_stat.f_iosize;
1573 int iomask = iosize - 1;
1578 * Align to iosize boundary, super-align to page boundary.
1581 cnt += off & iomask;
1582 off &= ~(u_quad_t)iomask;
1584 if (off & PAGE_MASK) {
1585 cnt += off & PAGE_MASK;
1586 off &= ~(u_quad_t)PAGE_MASK;
1588 lblkno = off / iosize;
1590 if (vp->v_object && vm_object_mightbedirty(vp->v_object)) {
1591 VM_OBJECT_WLOCK(vp->v_object);
1592 vm_object_page_clean(vp->v_object, off, off + cnt,
1594 VM_OBJECT_WUNLOCK(vp->v_object);
1603 * If we have a buffer and it is marked B_DELWRI we
1604 * have to lock and write it. Otherwise the prior
1605 * write is assumed to have already been committed.
1607 * gbincore() can return invalid buffers now so we
1608 * have to check that bit as well (though B_DELWRI
1609 * should not be set if B_INVAL is set there could be
1610 * a race here since we haven't locked the buffer).
1612 if ((bp = gbincore(&vp->v_bufobj, lblkno)) != NULL) {
1613 if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
1614 LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) {
1616 continue; /* retry */
1618 if ((bp->b_flags & (B_DELWRI|B_INVAL)) ==
1621 bp->b_flags &= ~B_ASYNC;
1644 nfsvno_statfs(struct vnode *vp, struct statfs *sf)
1650 if (nfsrv_devidcnt > 0) {
1651 /* For a pNFS service, get the DS numbers. */
1652 tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK | M_ZERO);
1653 error = nfsrv_pnfsstatfs(tsf, vp->v_mount);
1659 error = VFS_STATFS(vp->v_mount, sf);
1662 sf->f_blocks = tsf->f_blocks;
1663 sf->f_bavail = tsf->f_bavail;
1664 sf->f_bfree = tsf->f_bfree;
1665 sf->f_bsize = tsf->f_bsize;
1668 * Since NFS handles these values as unsigned on the
1669 * wire, there is no way to represent negative values,
1670 * so set them to 0. Without this, they will appear
1671 * to be very large positive values for clients like
1674 if (sf->f_bavail < 0)
1676 if (sf->f_ffree < 0)
1685 * Do the vnode op stuff for Open. Similar to nfsvno_createsub(), but
1686 * must handle nfsrv_opencheck() calls after any other access checks.
1689 nfsvno_open(struct nfsrv_descript *nd, struct nameidata *ndp,
1690 nfsquad_t clientid, nfsv4stateid_t *stateidp, struct nfsstate *stp,
1691 int *exclusive_flagp, struct nfsvattr *nvap, int32_t *cverf, int create,
1692 NFSACL_T *aclp, nfsattrbit_t *attrbitp, struct ucred *cred,
1693 struct nfsexstuff *exp, struct vnode **vpp)
1695 struct vnode *vp = NULL;
1697 struct nfsexstuff nes;
1698 struct thread *p = curthread;
1700 if (ndp->ni_vp == NULL)
1701 nd->nd_repstat = nfsrv_opencheck(clientid,
1702 stateidp, stp, NULL, nd, p, nd->nd_repstat);
1703 if (!nd->nd_repstat) {
1704 if (ndp->ni_vp == NULL) {
1705 vrele(ndp->ni_startdir);
1706 nd->nd_repstat = VOP_CREATE(ndp->ni_dvp,
1707 &ndp->ni_vp, &ndp->ni_cnd, &nvap->na_vattr);
1708 /* For a pNFS server, create the data file on a DS. */
1709 if (nd->nd_repstat == 0) {
1711 * Create a data file on a DS for a pNFS server.
1712 * This function just returns if not
1713 * running a pNFS DS or the creation fails.
1715 nfsrv_pnfscreate(ndp->ni_vp, &nvap->na_vattr,
1719 nfsvno_relpathbuf(ndp);
1720 if (!nd->nd_repstat) {
1721 if (*exclusive_flagp) {
1722 *exclusive_flagp = 0;
1723 NFSVNO_ATTRINIT(nvap);
1724 nvap->na_atime.tv_sec = cverf[0];
1725 nvap->na_atime.tv_nsec = cverf[1];
1726 nd->nd_repstat = VOP_SETATTR(ndp->ni_vp,
1727 &nvap->na_vattr, cred);
1728 if (nd->nd_repstat != 0) {
1731 nd->nd_repstat = NFSERR_NOTSUPP;
1733 NFSSETBIT_ATTRBIT(attrbitp,
1734 NFSATTRBIT_TIMEACCESS);
1736 nfsrv_fixattr(nd, ndp->ni_vp, nvap,
1737 aclp, p, attrbitp, exp);
1742 if (ndp->ni_startdir)
1743 vrele(ndp->ni_startdir);
1744 nfsvno_relpathbuf(ndp);
1746 if (create == NFSV4OPEN_CREATE) {
1747 if (ndp->ni_dvp == vp)
1752 if (NFSVNO_ISSETSIZE(nvap) && vp->v_type == VREG) {
1753 if (ndp->ni_cnd.cn_flags & RDONLY)
1754 NFSVNO_SETEXRDONLY(&nes);
1756 NFSVNO_EXINIT(&nes);
1757 nd->nd_repstat = nfsvno_accchk(vp,
1758 VWRITE, cred, &nes, p,
1759 NFSACCCHK_NOOVERRIDE,
1760 NFSACCCHK_VPISLOCKED, NULL);
1761 nd->nd_repstat = nfsrv_opencheck(clientid,
1762 stateidp, stp, vp, nd, p, nd->nd_repstat);
1763 if (!nd->nd_repstat) {
1764 tempsize = nvap->na_size;
1765 NFSVNO_ATTRINIT(nvap);
1766 nvap->na_size = tempsize;
1767 nd->nd_repstat = VOP_SETATTR(vp,
1768 &nvap->na_vattr, cred);
1770 } else if (vp->v_type == VREG) {
1771 nd->nd_repstat = nfsrv_opencheck(clientid,
1772 stateidp, stp, vp, nd, p, nd->nd_repstat);
1776 if (ndp->ni_cnd.cn_flags & HASBUF)
1777 nfsvno_relpathbuf(ndp);
1778 if (ndp->ni_startdir && create == NFSV4OPEN_CREATE) {
1779 vrele(ndp->ni_startdir);
1780 if (ndp->ni_dvp == ndp->ni_vp)
1790 NFSEXITCODE2(0, nd);
1794 * Updates the file rev and sets the mtime and ctime
1795 * to the current clock time, returning the va_filerev and va_Xtime
1797 * Return ESTALE to indicate the vnode is VIRF_DOOMED.
1800 nfsvno_updfilerev(struct vnode *vp, struct nfsvattr *nvap,
1801 struct nfsrv_descript *nd, struct thread *p)
1806 vfs_timestamp(&va.va_mtime);
1807 if (NFSVOPISLOCKED(vp) != LK_EXCLUSIVE) {
1808 NFSVOPLOCK(vp, LK_UPGRADE | LK_RETRY);
1809 if (VN_IS_DOOMED(vp))
1812 (void) VOP_SETATTR(vp, &va, nd->nd_cred);
1813 (void) nfsvno_getattr(vp, nvap, nd, p, 1, NULL);
1818 * Glue routine to nfsv4_fillattr().
1821 nfsvno_fillattr(struct nfsrv_descript *nd, struct mount *mp, struct vnode *vp,
1822 struct nfsvattr *nvap, fhandle_t *fhp, int rderror, nfsattrbit_t *attrbitp,
1823 struct ucred *cred, struct thread *p, int isdgram, int reterr,
1824 int supports_nfsv4acls, int at_root, uint64_t mounted_on_fileno)
1830 if (nfsrv_devidcnt > 0 &&
1831 (NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEAVAIL) ||
1832 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACEFREE) ||
1833 NFSISSET_ATTRBIT(attrbitp, NFSATTRBIT_SPACETOTAL))) {
1834 sf = malloc(sizeof(*sf), M_TEMP, M_WAITOK | M_ZERO);
1835 error = nfsrv_pnfsstatfs(sf, mp);
1841 error = nfsv4_fillattr(nd, mp, vp, NULL, &nvap->na_vattr, fhp, rderror,
1842 attrbitp, cred, p, isdgram, reterr, supports_nfsv4acls, at_root,
1843 mounted_on_fileno, sf);
1845 NFSEXITCODE2(0, nd);
1849 /* Since the Readdir vnode ops vary, put the entire functions in here. */
1851 * nfs readdir service
1852 * - mallocs what it thinks is enough to read
1853 * count rounded up to a multiple of DIRBLKSIZ <= NFS_MAXREADDIR
1854 * - calls VOP_READDIR()
1855 * - loops around building the reply
1856 * if the output generated exceeds count break out of loop
1857 * The NFSM_CLGET macro is used here so that the reply will be packed
1858 * tightly in mbuf clusters.
1859 * - it trims out records with d_fileno == 0
1860 * this doesn't matter for Unix clients, but they might confuse clients
1862 * - it trims out records with d_type == DT_WHT
1863 * these cannot be seen through NFS (unless we extend the protocol)
1864 * The alternate call nfsrvd_readdirplus() does lookups as well.
1865 * PS: The NFS protocol spec. does not clarify what the "count" byte
1866 * argument is a count of.. just name strings and file id's or the
1867 * entire reply rpc or ...
1868 * I tried just file name and id sizes and it confused the Sun client,
1869 * so I am using the full rpc size now. The "paranoia.." comment refers
1870 * to including the status longwords that are not a part of the dir.
1871 * "entry" structures, but are in the rpc.
1874 nfsrvd_readdir(struct nfsrv_descript *nd, int isdgram,
1875 struct vnode *vp, struct nfsexstuff *exp)
1880 char *cpos, *cend, *rbuf;
1882 int nlen, error = 0, getret = 1;
1883 int siz, cnt, fullsiz, eofflag, ncookies;
1884 u_int64_t off, toff, verf __unused;
1885 u_long *cookies = NULL, *cookiep;
1889 struct thread *p = curthread;
1891 if (nd->nd_repstat) {
1892 nfsrv_postopattr(nd, getret, &at);
1895 if (nd->nd_flag & ND_NFSV2) {
1896 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1897 off = fxdr_unsigned(u_quad_t, *tl++);
1899 NFSM_DISSECT(tl, u_int32_t *, 5 * NFSX_UNSIGNED);
1900 off = fxdr_hyper(tl);
1902 verf = fxdr_hyper(tl);
1906 cnt = fxdr_unsigned(int, *tl);
1907 if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
1908 cnt = NFS_SRVMAXDATA(nd);
1909 siz = ((cnt + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
1911 if (nd->nd_flag & ND_NFSV3) {
1912 nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1,
1916 * va_filerev is not sufficient as a cookie verifier,
1917 * since it is not supposed to change when entries are
1918 * removed/added unless that offset cookies returned to
1919 * the client are no longer valid.
1921 if (!nd->nd_repstat && toff && verf != at.na_filerev)
1922 nd->nd_repstat = NFSERR_BAD_COOKIE;
1925 if (!nd->nd_repstat && vp->v_type != VDIR)
1926 nd->nd_repstat = NFSERR_NOTDIR;
1927 if (nd->nd_repstat == 0 && cnt == 0) {
1928 if (nd->nd_flag & ND_NFSV2)
1929 /* NFSv2 does not have NFSERR_TOOSMALL */
1930 nd->nd_repstat = EPERM;
1932 nd->nd_repstat = NFSERR_TOOSMALL;
1934 if (!nd->nd_repstat)
1935 nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
1936 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
1937 NFSACCCHK_VPISLOCKED, NULL);
1938 if (nd->nd_repstat) {
1940 if (nd->nd_flag & ND_NFSV3)
1941 nfsrv_postopattr(nd, getret, &at);
1944 is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
1945 rbuf = malloc(siz, M_TEMP, M_WAITOK);
1949 free(cookies, M_TEMP);
1957 io.uio_offset = (off_t)off;
1959 io.uio_segflg = UIO_SYSSPACE;
1960 io.uio_rw = UIO_READ;
1962 nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
1964 off = (u_int64_t)io.uio_offset;
1966 siz -= io.uio_resid;
1968 if (!cookies && !nd->nd_repstat)
1969 nd->nd_repstat = NFSERR_PERM;
1970 if (nd->nd_flag & ND_NFSV3) {
1971 getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
1972 if (!nd->nd_repstat)
1973 nd->nd_repstat = getret;
1977 * Handles the failed cases. nd->nd_repstat == 0 past here.
1979 if (nd->nd_repstat) {
1983 free(cookies, M_TEMP);
1984 if (nd->nd_flag & ND_NFSV3)
1985 nfsrv_postopattr(nd, getret, &at);
1989 * If nothing read, return eof
1994 if (nd->nd_flag & ND_NFSV2) {
1995 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
1997 nfsrv_postopattr(nd, getret, &at);
1998 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
1999 txdr_hyper(at.na_filerev, tl);
2002 *tl++ = newnfs_false;
2005 free(cookies, M_TEMP);
2010 * Check for degenerate cases of nothing useful read.
2011 * If so go try again
2015 dp = (struct dirent *)cpos;
2019 * For some reason FreeBSD's ufs_readdir() chooses to back the
2020 * directory offset up to a block boundary, so it is necessary to
2021 * skip over the records that precede the requested offset. This
2022 * requires the assumption that file offset cookies monotonically
2025 while (cpos < cend && ncookies > 0 &&
2026 (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
2027 (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff))) {
2028 cpos += dp->d_reclen;
2029 dp = (struct dirent *)cpos;
2033 if (cpos >= cend || ncookies == 0) {
2041 * dirlen is the size of the reply, including all XDR and must
2042 * not exceed cnt. For NFSv2, RFC1094 didn't clearly indicate
2043 * if the XDR should be included in "count", but to be safe, we do.
2044 * (Include the two booleans at the end of the reply in dirlen now.)
2046 if (nd->nd_flag & ND_NFSV3) {
2047 nfsrv_postopattr(nd, getret, &at);
2048 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2049 txdr_hyper(at.na_filerev, tl);
2050 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
2052 dirlen = 2 * NFSX_UNSIGNED;
2055 /* Loop through the records and build reply */
2056 while (cpos < cend && ncookies > 0) {
2057 nlen = dp->d_namlen;
2058 if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2059 nlen <= NFS_MAXNAMLEN) {
2060 if (nd->nd_flag & ND_NFSV3)
2061 dirlen += (6*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
2063 dirlen += (4*NFSX_UNSIGNED + NFSM_RNDUP(nlen));
2070 * Build the directory record xdr from
2073 if (nd->nd_flag & ND_NFSV3) {
2074 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2075 *tl++ = newnfs_true;
2078 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2079 *tl++ = newnfs_true;
2081 *tl = txdr_unsigned(dp->d_fileno);
2082 (void) nfsm_strtom(nd, dp->d_name, nlen);
2083 if (nd->nd_flag & ND_NFSV3) {
2084 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2087 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
2088 *tl = txdr_unsigned(*cookiep);
2090 cpos += dp->d_reclen;
2091 dp = (struct dirent *)cpos;
2097 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2098 *tl++ = newnfs_false;
2104 free(cookies, M_TEMP);
2107 NFSEXITCODE2(0, nd);
2111 NFSEXITCODE2(error, nd);
2116 * Readdirplus for V3 and Readdir for V4.
2119 nfsrvd_readdirplus(struct nfsrv_descript *nd, int isdgram,
2120 struct vnode *vp, struct nfsexstuff *exp)
2125 char *cpos, *cend, *rbuf;
2128 struct nfsvattr nva, at, *nvap = &nva;
2129 struct mbuf *mb0, *mb1;
2130 struct nfsreferral *refp;
2131 int nlen, r, error = 0, getret = 1, usevget = 1;
2132 int siz, cnt, fullsiz, eofflag, ncookies, entrycnt;
2133 caddr_t bpos0, bpos1;
2134 u_int64_t off, toff, verf;
2135 u_long *cookies = NULL, *cookiep;
2136 nfsattrbit_t attrbits, rderrbits, savbits;
2139 struct componentname cn;
2140 int at_root, is_ufs, is_zfs, needs_unbusy, supports_nfsv4acls;
2141 struct mount *mp, *new_mp;
2142 uint64_t mounted_on_fileno;
2143 struct thread *p = curthread;
2145 if (nd->nd_repstat) {
2146 nfsrv_postopattr(nd, getret, &at);
2149 NFSM_DISSECT(tl, u_int32_t *, 6 * NFSX_UNSIGNED);
2150 off = fxdr_hyper(tl);
2153 verf = fxdr_hyper(tl);
2155 siz = fxdr_unsigned(int, *tl++);
2156 cnt = fxdr_unsigned(int, *tl);
2159 * Use the server's maximum data transfer size as the upper bound
2162 if (cnt > NFS_SRVMAXDATA(nd) || cnt < 0)
2163 cnt = NFS_SRVMAXDATA(nd);
2166 * siz is a "hint" of how much directory information (name, fileid,
2167 * cookie) should be in the reply. At least one client "hints" 0,
2168 * so I set it to cnt for that case. I also round it up to the
2169 * next multiple of DIRBLKSIZ.
2170 * Since the size of a Readdirplus directory entry reply will always
2171 * be greater than a directory entry returned by VOP_READDIR(), it
2172 * does not make sense to read more than NFS_SRVMAXDATA() via
2177 else if (siz > NFS_SRVMAXDATA(nd))
2178 siz = NFS_SRVMAXDATA(nd);
2179 siz = ((siz + DIRBLKSIZ - 1) & ~(DIRBLKSIZ - 1));
2181 if (nd->nd_flag & ND_NFSV4) {
2182 error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
2185 NFSSET_ATTRBIT(&savbits, &attrbits);
2186 NFSCLRNOTFILLABLE_ATTRBIT(&attrbits, nd);
2187 NFSZERO_ATTRBIT(&rderrbits);
2188 NFSSETBIT_ATTRBIT(&rderrbits, NFSATTRBIT_RDATTRERROR);
2190 NFSZERO_ATTRBIT(&attrbits);
2193 nd->nd_repstat = getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
2195 if (!nd->nd_repstat) {
2196 if (off && verf != at.na_filerev) {
2198 * va_filerev is not sufficient as a cookie verifier,
2199 * since it is not supposed to change when entries are
2200 * removed/added unless that offset cookies returned to
2201 * the client are no longer valid.
2203 if (nd->nd_flag & ND_NFSV4) {
2204 nd->nd_repstat = NFSERR_NOTSAME;
2206 nd->nd_repstat = NFSERR_BAD_COOKIE;
2211 if (!nd->nd_repstat && vp->v_type != VDIR)
2212 nd->nd_repstat = NFSERR_NOTDIR;
2213 if (!nd->nd_repstat && cnt == 0)
2214 nd->nd_repstat = NFSERR_TOOSMALL;
2215 if (!nd->nd_repstat)
2216 nd->nd_repstat = nfsvno_accchk(vp, VEXEC,
2217 nd->nd_cred, exp, p, NFSACCCHK_NOOVERRIDE,
2218 NFSACCCHK_VPISLOCKED, NULL);
2219 if (nd->nd_repstat) {
2221 if (nd->nd_flag & ND_NFSV3)
2222 nfsrv_postopattr(nd, getret, &at);
2225 is_ufs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "ufs") == 0;
2226 is_zfs = strcmp(vp->v_mount->mnt_vfc->vfc_name, "zfs") == 0;
2228 rbuf = malloc(siz, M_TEMP, M_WAITOK);
2232 free(cookies, M_TEMP);
2240 io.uio_offset = (off_t)off;
2242 io.uio_segflg = UIO_SYSSPACE;
2243 io.uio_rw = UIO_READ;
2245 nd->nd_repstat = VOP_READDIR(vp, &io, nd->nd_cred, &eofflag, &ncookies,
2247 off = (u_int64_t)io.uio_offset;
2249 siz -= io.uio_resid;
2251 getret = nfsvno_getattr(vp, &at, nd, p, 1, NULL);
2253 if (!cookies && !nd->nd_repstat)
2254 nd->nd_repstat = NFSERR_PERM;
2255 if (!nd->nd_repstat)
2256 nd->nd_repstat = getret;
2257 if (nd->nd_repstat) {
2260 free(cookies, M_TEMP);
2262 if (nd->nd_flag & ND_NFSV3)
2263 nfsrv_postopattr(nd, getret, &at);
2267 * If nothing read, return eof
2272 if (nd->nd_flag & ND_NFSV3)
2273 nfsrv_postopattr(nd, getret, &at);
2274 NFSM_BUILD(tl, u_int32_t *, 4 * NFSX_UNSIGNED);
2275 txdr_hyper(at.na_filerev, tl);
2277 *tl++ = newnfs_false;
2279 free(cookies, M_TEMP);
2285 * Check for degenerate cases of nothing useful read.
2286 * If so go try again
2290 dp = (struct dirent *)cpos;
2294 * For some reason FreeBSD's ufs_readdir() chooses to back the
2295 * directory offset up to a block boundary, so it is necessary to
2296 * skip over the records that precede the requested offset. This
2297 * requires the assumption that file offset cookies monotonically
2300 while (cpos < cend && ncookies > 0 &&
2301 (dp->d_fileno == 0 || dp->d_type == DT_WHT ||
2302 (is_ufs == 1 && ((u_quad_t)(*cookiep)) <= toff) ||
2303 ((nd->nd_flag & ND_NFSV4) &&
2304 ((dp->d_namlen == 1 && dp->d_name[0] == '.') ||
2305 (dp->d_namlen==2 && dp->d_name[0]=='.' && dp->d_name[1]=='.'))))) {
2306 cpos += dp->d_reclen;
2307 dp = (struct dirent *)cpos;
2311 if (cpos >= cend || ncookies == 0) {
2318 * Busy the file system so that the mount point won't go away
2319 * and, as such, VFS_VGET() can be used safely.
2324 nd->nd_repstat = vfs_busy(mp, 0);
2326 if (nd->nd_repstat != 0) {
2328 free(cookies, M_TEMP);
2330 if (nd->nd_flag & ND_NFSV3)
2331 nfsrv_postopattr(nd, getret, &at);
2336 * Check to see if entries in this directory can be safely acquired
2337 * via VFS_VGET() or if a switch to VOP_LOOKUP() is required.
2338 * ZFS snapshot directories need VOP_LOOKUP(), so that any
2339 * automount of the snapshot directory that is required will
2341 * This needs to be done here for NFSv4, since NFSv4 never does
2342 * a VFS_VGET() for "." or "..".
2345 r = VFS_VGET(mp, at.na_fileid, LK_SHARED, &nvp);
2346 if (r == EOPNOTSUPP) {
2348 cn.cn_nameiop = LOOKUP;
2349 cn.cn_lkflags = LK_SHARED | LK_RETRY;
2350 cn.cn_cred = nd->nd_cred;
2357 * Save this position, in case there is an error before one entry
2361 bpos0 = nd->nd_bpos;
2364 * Fill in the first part of the reply.
2365 * dirlen is the reply length in bytes and cannot exceed cnt.
2366 * (Include the two booleans at the end of the reply in dirlen now,
2367 * so we recognize when we have exceeded cnt.)
2369 if (nd->nd_flag & ND_NFSV3) {
2370 dirlen = NFSX_V3POSTOPATTR + NFSX_VERF + 2 * NFSX_UNSIGNED;
2371 nfsrv_postopattr(nd, getret, &at);
2373 dirlen = NFSX_VERF + 2 * NFSX_UNSIGNED;
2375 NFSM_BUILD(tl, u_int32_t *, NFSX_VERF);
2376 txdr_hyper(at.na_filerev, tl);
2379 * Save this position, in case there is an empty reply needed.
2382 bpos1 = nd->nd_bpos;
2384 /* Loop through the records and build reply */
2386 while (cpos < cend && ncookies > 0 && dirlen < cnt) {
2387 nlen = dp->d_namlen;
2388 if (dp->d_fileno != 0 && dp->d_type != DT_WHT &&
2389 nlen <= NFS_MAXNAMLEN &&
2390 ((nd->nd_flag & ND_NFSV3) || nlen > 2 ||
2391 (nlen==2 && (dp->d_name[0]!='.' || dp->d_name[1]!='.'))
2392 || (nlen == 1 && dp->d_name[0] != '.'))) {
2394 * Save the current position in the reply, in case
2395 * this entry exceeds cnt.
2398 bpos1 = nd->nd_bpos;
2401 * For readdir_and_lookup get the vnode using
2410 mounted_on_fileno = (uint64_t)dp->d_fileno;
2411 if ((nd->nd_flag & ND_NFSV3) ||
2412 NFSNONZERO_ATTRBIT(&savbits)) {
2413 if (nd->nd_flag & ND_NFSV4)
2414 refp = nfsv4root_getreferral(NULL,
2418 r = VFS_VGET(mp, dp->d_fileno,
2422 if (r == EOPNOTSUPP) {
2425 cn.cn_nameiop = LOOKUP;
2433 cn.cn_nameptr = dp->d_name;
2434 cn.cn_namelen = nlen;
2435 cn.cn_flags = ISLASTCN |
2436 NOFOLLOW | LOCKLEAF;
2438 dp->d_name[0] == '.' &&
2439 dp->d_name[1] == '.')
2442 if (NFSVOPLOCK(vp, LK_SHARED)
2444 nd->nd_repstat = EPERM;
2447 if ((vp->v_vflag & VV_ROOT) != 0
2448 && (cn.cn_flags & ISDOTDOT)
2454 r = VOP_LOOKUP(vp, &nvp,
2462 * For NFSv4, check to see if nvp is
2463 * a mount point and get the mount
2464 * point vnode, as required.
2467 nfsrv_enable_crossmntpt != 0 &&
2468 (nd->nd_flag & ND_NFSV4) != 0 &&
2469 nvp->v_type == VDIR &&
2470 nvp->v_mountedhere != NULL) {
2471 new_mp = nvp->v_mountedhere;
2472 r = vfs_busy(new_mp, 0);
2476 r = VFS_ROOT(new_mp,
2486 * If we failed to look up the entry, then it
2487 * has become invalid, most likely removed.
2494 KASSERT(refp != NULL || nvp != NULL,
2495 ("%s: undetected lookup error", __func__));
2498 ((nd->nd_flag & ND_NFSV3) ||
2499 NFSNONZERO_ATTRBIT(&attrbits))) {
2500 r = nfsvno_getfh(nvp, &nfh, p);
2502 r = nfsvno_getattr(nvp, nvap, nd, p,
2504 if (r == 0 && is_zfs == 1 &&
2505 nfsrv_enable_crossmntpt != 0 &&
2506 (nd->nd_flag & ND_NFSV4) != 0 &&
2507 nvp->v_type == VDIR &&
2508 vp->v_mount != nvp->v_mount) {
2510 * For a ZFS snapshot, there is a
2511 * pseudo mount that does not set
2512 * v_mountedhere, so it needs to
2513 * be detected via a different
2518 new_mp = nvp->v_mount;
2523 * If we failed to get attributes of the entry,
2524 * then just skip it for NFSv3 (the traditional
2525 * behavior in the old NFS server).
2526 * For NFSv4 the behavior is controlled by
2527 * RDATTRERROR: we either ignore the error or
2529 * Note that RDATTRERROR is never set for NFSv3.
2532 if (!NFSISSET_ATTRBIT(&attrbits,
2533 NFSATTRBIT_RDATTRERROR)) {
2535 if (needs_unbusy != 0)
2537 if ((nd->nd_flag & ND_NFSV3))
2546 * Build the directory record xdr
2548 if (nd->nd_flag & ND_NFSV3) {
2549 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2550 *tl++ = newnfs_true;
2552 *tl = txdr_unsigned(dp->d_fileno);
2553 dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2554 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2556 *tl = txdr_unsigned(*cookiep);
2557 nfsrv_postopattr(nd, 0, nvap);
2558 dirlen += nfsm_fhtom(nd,(u_int8_t *)&nfh,0,1);
2559 dirlen += (5*NFSX_UNSIGNED+NFSX_V3POSTOPATTR);
2563 NFSM_BUILD(tl, u_int32_t *, 3 * NFSX_UNSIGNED);
2564 *tl++ = newnfs_true;
2566 *tl = txdr_unsigned(*cookiep);
2567 dirlen += nfsm_strtom(nd, dp->d_name, nlen);
2569 supports_nfsv4acls =
2570 nfs_supportsnfsv4acls(nvp);
2573 supports_nfsv4acls = 0;
2575 dirlen += nfsrv_putreferralattr(nd,
2578 if (nd->nd_repstat) {
2581 if (needs_unbusy != 0)
2586 dirlen += nfsvno_fillattr(nd, new_mp,
2587 nvp, nvap, &nfh, r, &rderrbits,
2588 nd->nd_cred, p, isdgram, 0,
2589 supports_nfsv4acls, at_root,
2592 dirlen += nfsvno_fillattr(nd, new_mp,
2593 nvp, nvap, &nfh, r, &attrbits,
2594 nd->nd_cred, p, isdgram, 0,
2595 supports_nfsv4acls, at_root,
2600 dirlen += (3 * NFSX_UNSIGNED);
2602 if (needs_unbusy != 0)
2608 cpos += dp->d_reclen;
2609 dp = (struct dirent *)cpos;
2617 * If dirlen > cnt, we must strip off the last entry. If that
2618 * results in an empty reply, report NFSERR_TOOSMALL.
2620 if (dirlen > cnt || nd->nd_repstat) {
2621 if (!nd->nd_repstat && entrycnt == 0)
2622 nd->nd_repstat = NFSERR_TOOSMALL;
2623 if (nd->nd_repstat) {
2624 newnfs_trimtrailing(nd, mb0, bpos0);
2625 if (nd->nd_flag & ND_NFSV3)
2626 nfsrv_postopattr(nd, getret, &at);
2628 newnfs_trimtrailing(nd, mb1, bpos1);
2630 } else if (cpos < cend)
2632 if (!nd->nd_repstat) {
2633 NFSM_BUILD(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2634 *tl++ = newnfs_false;
2640 free(cookies, M_TEMP);
2644 NFSEXITCODE2(0, nd);
2648 NFSEXITCODE2(error, nd);
2653 * Get the settable attributes out of the mbuf list.
2654 * (Return 0 or EBADRPC)
2657 nfsrv_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap,
2658 nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2661 struct nfsv2_sattr *sp;
2662 int error = 0, toclient = 0;
2664 switch (nd->nd_flag & (ND_NFSV2 | ND_NFSV3 | ND_NFSV4)) {
2666 NFSM_DISSECT(sp, struct nfsv2_sattr *, NFSX_V2SATTR);
2668 * Some old clients didn't fill in the high order 16bits.
2669 * --> check the low order 2 bytes for 0xffff
2671 if ((fxdr_unsigned(int, sp->sa_mode) & 0xffff) != 0xffff)
2672 nvap->na_mode = nfstov_mode(sp->sa_mode);
2673 if (sp->sa_uid != newnfs_xdrneg1)
2674 nvap->na_uid = fxdr_unsigned(uid_t, sp->sa_uid);
2675 if (sp->sa_gid != newnfs_xdrneg1)
2676 nvap->na_gid = fxdr_unsigned(gid_t, sp->sa_gid);
2677 if (sp->sa_size != newnfs_xdrneg1)
2678 nvap->na_size = fxdr_unsigned(u_quad_t, sp->sa_size);
2679 if (sp->sa_atime.nfsv2_sec != newnfs_xdrneg1) {
2681 fxdr_nfsv2time(&sp->sa_atime, &nvap->na_atime);
2683 nvap->na_atime.tv_sec =
2684 fxdr_unsigned(u_int32_t,sp->sa_atime.nfsv2_sec);
2685 nvap->na_atime.tv_nsec = 0;
2688 if (sp->sa_mtime.nfsv2_sec != newnfs_xdrneg1)
2689 fxdr_nfsv2time(&sp->sa_mtime, &nvap->na_mtime);
2692 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2693 if (*tl == newnfs_true) {
2694 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2695 nvap->na_mode = nfstov_mode(*tl);
2697 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2698 if (*tl == newnfs_true) {
2699 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2700 nvap->na_uid = fxdr_unsigned(uid_t, *tl);
2702 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2703 if (*tl == newnfs_true) {
2704 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2705 nvap->na_gid = fxdr_unsigned(gid_t, *tl);
2707 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2708 if (*tl == newnfs_true) {
2709 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2710 nvap->na_size = fxdr_hyper(tl);
2712 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2713 switch (fxdr_unsigned(int, *tl)) {
2714 case NFSV3SATTRTIME_TOCLIENT:
2715 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2716 fxdr_nfsv3time(tl, &nvap->na_atime);
2719 case NFSV3SATTRTIME_TOSERVER:
2720 vfs_timestamp(&nvap->na_atime);
2721 nvap->na_vaflags |= VA_UTIMES_NULL;
2724 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2725 switch (fxdr_unsigned(int, *tl)) {
2726 case NFSV3SATTRTIME_TOCLIENT:
2727 NFSM_DISSECT(tl, u_int32_t *, 2 * NFSX_UNSIGNED);
2728 fxdr_nfsv3time(tl, &nvap->na_mtime);
2729 nvap->na_vaflags &= ~VA_UTIMES_NULL;
2731 case NFSV3SATTRTIME_TOSERVER:
2732 vfs_timestamp(&nvap->na_mtime);
2734 nvap->na_vaflags |= VA_UTIMES_NULL;
2739 error = nfsv4_sattr(nd, vp, nvap, attrbitp, aclp, p);
2742 NFSEXITCODE2(error, nd);
2747 * Handle the setable attributes for V4.
2748 * Returns NFSERR_BADXDR if it can't be parsed, 0 otherwise.
2751 nfsv4_sattr(struct nfsrv_descript *nd, vnode_t vp, struct nfsvattr *nvap,
2752 nfsattrbit_t *attrbitp, NFSACL_T *aclp, struct thread *p)
2757 int error, attrsize, bitpos, aclsize, aceerr, retnotsup = 0;
2758 int moderet, toclient = 0;
2759 u_char *cp, namestr[NFSV4_SMALLSTR + 1];
2762 u_short mode, mask; /* Same type as va_mode. */
2765 error = nfsrv_getattrbits(nd, attrbitp, NULL, &retnotsup);
2768 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2769 attrsize = fxdr_unsigned(int, *tl);
2772 * Loop around getting the setable attributes. If an unsupported
2773 * one is found, set nd_repstat == NFSERR_ATTRNOTSUPP and return.
2776 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2777 bitpos = NFSATTRBIT_MAX;
2782 for (; bitpos < NFSATTRBIT_MAX; bitpos++) {
2783 if (attrsum > attrsize) {
2784 error = NFSERR_BADXDR;
2787 if (NFSISSET_ATTRBIT(attrbitp, bitpos))
2789 case NFSATTRBIT_SIZE:
2790 NFSM_DISSECT(tl, u_int32_t *, NFSX_HYPER);
2791 if (vp != NULL && vp->v_type != VREG) {
2792 error = (vp->v_type == VDIR) ? NFSERR_ISDIR :
2796 nvap->na_size = fxdr_hyper(tl);
2797 attrsum += NFSX_HYPER;
2799 case NFSATTRBIT_ACL:
2800 error = nfsrv_dissectacl(nd, aclp, &aceerr, &aclsize,
2804 if (aceerr && !nd->nd_repstat)
2805 nd->nd_repstat = aceerr;
2808 case NFSATTRBIT_ARCHIVE:
2809 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2810 if (!nd->nd_repstat)
2811 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2812 attrsum += NFSX_UNSIGNED;
2814 case NFSATTRBIT_HIDDEN:
2815 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2816 if (!nd->nd_repstat)
2817 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2818 attrsum += NFSX_UNSIGNED;
2820 case NFSATTRBIT_MIMETYPE:
2821 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2822 i = fxdr_unsigned(int, *tl);
2823 error = nfsm_advance(nd, NFSM_RNDUP(i), -1);
2826 if (!nd->nd_repstat)
2827 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2828 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(i));
2830 case NFSATTRBIT_MODE:
2831 moderet = NFSERR_INVAL; /* Can't do MODESETMASKED. */
2832 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2833 nvap->na_mode = nfstov_mode(*tl);
2834 attrsum += NFSX_UNSIGNED;
2836 case NFSATTRBIT_OWNER:
2837 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2838 j = fxdr_unsigned(int, *tl);
2840 error = NFSERR_BADXDR;
2843 if (j > NFSV4_SMALLSTR)
2844 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2847 error = nfsrv_mtostr(nd, cp, j);
2849 if (j > NFSV4_SMALLSTR)
2850 free(cp, M_NFSSTRING);
2853 if (!nd->nd_repstat) {
2854 nd->nd_repstat = nfsv4_strtouid(nd, cp, j,
2856 if (!nd->nd_repstat)
2859 if (j > NFSV4_SMALLSTR)
2860 free(cp, M_NFSSTRING);
2861 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2863 case NFSATTRBIT_OWNERGROUP:
2864 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2865 j = fxdr_unsigned(int, *tl);
2867 error = NFSERR_BADXDR;
2870 if (j > NFSV4_SMALLSTR)
2871 cp = malloc(j + 1, M_NFSSTRING, M_WAITOK);
2874 error = nfsrv_mtostr(nd, cp, j);
2876 if (j > NFSV4_SMALLSTR)
2877 free(cp, M_NFSSTRING);
2880 if (!nd->nd_repstat) {
2881 nd->nd_repstat = nfsv4_strtogid(nd, cp, j,
2883 if (!nd->nd_repstat)
2886 if (j > NFSV4_SMALLSTR)
2887 free(cp, M_NFSSTRING);
2888 attrsum += (NFSX_UNSIGNED + NFSM_RNDUP(j));
2890 case NFSATTRBIT_SYSTEM:
2891 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2892 if (!nd->nd_repstat)
2893 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2894 attrsum += NFSX_UNSIGNED;
2896 case NFSATTRBIT_TIMEACCESSSET:
2897 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2898 attrsum += NFSX_UNSIGNED;
2899 if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2900 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2901 fxdr_nfsv4time(tl, &nvap->na_atime);
2903 attrsum += NFSX_V4TIME;
2905 vfs_timestamp(&nvap->na_atime);
2906 nvap->na_vaflags |= VA_UTIMES_NULL;
2909 case NFSATTRBIT_TIMEBACKUP:
2910 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2911 if (!nd->nd_repstat)
2912 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2913 attrsum += NFSX_V4TIME;
2915 case NFSATTRBIT_TIMECREATE:
2916 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2917 if (!nd->nd_repstat)
2918 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2919 attrsum += NFSX_V4TIME;
2921 case NFSATTRBIT_TIMEMODIFYSET:
2922 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
2923 attrsum += NFSX_UNSIGNED;
2924 if (fxdr_unsigned(int, *tl)==NFSV4SATTRTIME_TOCLIENT) {
2925 NFSM_DISSECT(tl, u_int32_t *, NFSX_V4TIME);
2926 fxdr_nfsv4time(tl, &nvap->na_mtime);
2927 nvap->na_vaflags &= ~VA_UTIMES_NULL;
2928 attrsum += NFSX_V4TIME;
2930 vfs_timestamp(&nvap->na_mtime);
2932 nvap->na_vaflags |= VA_UTIMES_NULL;
2935 case NFSATTRBIT_MODESETMASKED:
2936 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
2937 mode = fxdr_unsigned(u_short, *tl++);
2938 mask = fxdr_unsigned(u_short, *tl);
2940 * vp == NULL implies an Open/Create operation.
2941 * This attribute can only be used for Setattr and
2942 * only for NFSv4.1 or higher.
2943 * If moderet != 0, a mode attribute has also been
2944 * specified and this attribute cannot be done in the
2945 * same Setattr operation.
2947 if ((nd->nd_flag & ND_NFSV41) == 0)
2948 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2949 else if ((mode & ~07777) != 0 || (mask & ~07777) != 0 ||
2951 nd->nd_repstat = NFSERR_INVAL;
2952 else if (moderet == 0)
2953 moderet = VOP_GETATTR(vp, &va, nd->nd_cred);
2955 nvap->na_mode = (mode & mask) |
2956 (va.va_mode & ~mask);
2958 nd->nd_repstat = moderet;
2959 attrsum += 2 * NFSX_UNSIGNED;
2962 nd->nd_repstat = NFSERR_ATTRNOTSUPP;
2964 * set bitpos so we drop out of the loop.
2966 bitpos = NFSATTRBIT_MAX;
2972 * some clients pad the attrlist, so we need to skip over the
2975 if (attrsum > attrsize) {
2976 error = NFSERR_BADXDR;
2978 attrsize = NFSM_RNDUP(attrsize);
2979 if (attrsum < attrsize)
2980 error = nfsm_advance(nd, attrsize - attrsum, -1);
2983 NFSEXITCODE2(error, nd);
2988 * Check/setup export credentials.
2991 nfsd_excred(struct nfsrv_descript *nd, struct nfsexstuff *exp,
2992 struct ucred *credanon)
2997 * Check/setup credentials.
2999 if (nd->nd_flag & ND_GSS)
3000 exp->nes_exflag &= ~MNT_EXPORTANON;
3003 * Check to see if the operation is allowed for this security flavor.
3004 * RFC2623 suggests that the NFSv3 Fsinfo RPC be allowed to
3005 * AUTH_NONE or AUTH_SYS for file systems requiring RPCSEC_GSS.
3006 * Also, allow Secinfo, so that it can acquire the correct flavor(s).
3008 if (nfsvno_testexp(nd, exp) &&
3009 nd->nd_procnum != NFSV4OP_SECINFO &&
3010 nd->nd_procnum != NFSPROC_FSINFO) {
3011 if (nd->nd_flag & ND_NFSV4)
3012 error = NFSERR_WRONGSEC;
3014 error = (NFSERR_AUTHERR | AUTH_TOOWEAK);
3019 * Check to see if the file system is exported V4 only.
3021 if (NFSVNO_EXV4ONLY(exp) && !(nd->nd_flag & ND_NFSV4)) {
3022 error = NFSERR_PROGNOTV4;
3027 * Now, map the user credentials.
3028 * (Note that ND_AUTHNONE will only be set for an NFSv3
3029 * Fsinfo RPC. If set for anything else, this code might need
3032 if (NFSVNO_EXPORTED(exp)) {
3033 if (((nd->nd_flag & ND_GSS) == 0 && nd->nd_cred->cr_uid == 0) ||
3034 NFSVNO_EXPORTANON(exp) ||
3035 (nd->nd_flag & ND_AUTHNONE) != 0) {
3036 nd->nd_cred->cr_uid = credanon->cr_uid;
3037 nd->nd_cred->cr_gid = credanon->cr_gid;
3038 crsetgroups(nd->nd_cred, credanon->cr_ngroups,
3039 credanon->cr_groups);
3040 } else if ((nd->nd_flag & ND_GSS) == 0) {
3042 * If using AUTH_SYS, call nfsrv_getgrpscred() to see
3043 * if there is a replacement credential with a group
3044 * list set up by "nfsuserd -manage-gids".
3045 * If there is no replacement, nfsrv_getgrpscred()
3046 * simply returns its argument.
3048 nd->nd_cred = nfsrv_getgrpscred(nd->nd_cred);
3053 NFSEXITCODE2(error, nd);
3061 nfsvno_checkexp(struct mount *mp, struct sockaddr *nam, struct nfsexstuff *exp,
3062 struct ucred **credp)
3066 error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
3067 &exp->nes_numsecflavor, exp->nes_secflavors);
3069 if (nfs_rootfhset) {
3070 exp->nes_exflag = 0;
3071 exp->nes_numsecflavor = 0;
3074 } else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor >
3076 printf("nfsvno_checkexp: numsecflavors out of range\n");
3077 exp->nes_numsecflavor = 0;
3085 * Get a vnode for a file handle and export stuff.
3088 nfsvno_fhtovp(struct mount *mp, fhandle_t *fhp, struct sockaddr *nam,
3089 int lktype, struct vnode **vpp, struct nfsexstuff *exp,
3090 struct ucred **credp)
3095 exp->nes_numsecflavor = 0;
3096 error = VFS_FHTOVP(mp, &fhp->fh_fid, lktype, vpp);
3098 /* Make sure the server replies ESTALE to the client. */
3100 if (nam && !error) {
3101 error = VFS_CHECKEXP(mp, nam, &exp->nes_exflag, credp,
3102 &exp->nes_numsecflavor, exp->nes_secflavors);
3104 if (nfs_rootfhset) {
3105 exp->nes_exflag = 0;
3106 exp->nes_numsecflavor = 0;
3111 } else if (exp->nes_numsecflavor < 1 || exp->nes_numsecflavor >
3113 printf("nfsvno_fhtovp: numsecflavors out of range\n");
3114 exp->nes_numsecflavor = 0;
3124 * nfsd_fhtovp() - convert a fh to a vnode ptr
3125 * - look up fsid in mount list (if not found ret error)
3126 * - get vp and export rights by calling nfsvno_fhtovp()
3127 * - if cred->cr_uid == 0 or MNT_EXPORTANON set it to credanon
3129 * - if mpp != NULL, return the mount point so that it can
3130 * be used for vn_finished_write() by the caller
3133 nfsd_fhtovp(struct nfsrv_descript *nd, struct nfsrvfh *nfp, int lktype,
3134 struct vnode **vpp, struct nfsexstuff *exp,
3135 struct mount **mpp, int startwrite)
3138 struct ucred *credanon;
3141 fhp = (fhandle_t *)nfp->nfsrvfh_data;
3143 * Check for the special case of the nfsv4root_fh.
3145 mp = vfs_busyfs(&fhp->fh_fsid);
3150 nd->nd_repstat = ESTALE;
3155 vn_start_write(NULL, mpp, V_WAIT);
3156 if (lktype == LK_SHARED && !(MNT_SHARED_WRITES(mp)))
3157 lktype = LK_EXCLUSIVE;
3159 nd->nd_repstat = nfsvno_fhtovp(mp, fhp, nd->nd_nam, lktype, vpp, exp,
3164 * For NFSv4 without a pseudo root fs, unexported file handles
3165 * can be returned, so that Lookup works everywhere.
3167 if (!nd->nd_repstat && exp->nes_exflag == 0 &&
3168 !(nd->nd_flag & ND_NFSV4)) {
3170 nd->nd_repstat = EACCES;
3174 * Personally, I've never seen any point in requiring a
3175 * reserved port#, since only in the rare case where the
3176 * clients are all boxes with secure system privileges,
3177 * does it provide any enhanced security, but... some people
3178 * believe it to be useful and keep putting this code back in.
3179 * (There is also some "security checker" out there that
3180 * complains if the nfs server doesn't enforce this.)
3181 * However, note the following:
3182 * RFC3530 (NFSv4) specifies that a reserved port# not be
3184 * RFC2623 recommends that, if a reserved port# is checked for,
3185 * that there be a way to turn that off--> ifdef'd.
3187 #ifdef NFS_REQRSVPORT
3188 if (!nd->nd_repstat) {
3189 struct sockaddr_in *saddr;
3190 struct sockaddr_in6 *saddr6;
3192 saddr = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in *);
3193 saddr6 = NFSSOCKADDR(nd->nd_nam, struct sockaddr_in6 *);
3194 if (!(nd->nd_flag & ND_NFSV4) &&
3195 ((saddr->sin_family == AF_INET &&
3196 ntohs(saddr->sin_port) >= IPPORT_RESERVED) ||
3197 (saddr6->sin6_family == AF_INET6 &&
3198 ntohs(saddr6->sin6_port) >= IPPORT_RESERVED))) {
3200 nd->nd_repstat = (NFSERR_AUTHERR | AUTH_TOOWEAK);
3203 #endif /* NFS_REQRSVPORT */
3206 * Check/setup credentials.
3208 if (!nd->nd_repstat) {
3209 nd->nd_saveduid = nd->nd_cred->cr_uid;
3210 nd->nd_repstat = nfsd_excred(nd, exp, credanon);
3214 if (credanon != NULL)
3216 if (nd->nd_repstat) {
3218 vn_finished_write(mp);
3225 NFSEXITCODE2(0, nd);
3232 fp_getfvp(struct thread *p, int fd, struct file **fpp, struct vnode **vpp)
3234 struct filedesc *fdp;
3238 fdp = p->td_proc->p_fd;
3239 if (fd < 0 || fd >= fdp->fd_nfiles ||
3240 (fp = fdp->fd_ofiles[fd].fde_file) == NULL) {
3252 * Called from nfssvc() to update the exports list. Just call
3253 * vfs_export(). This has to be done, since the v4 root fake fs isn't
3254 * in the mount list.
3257 nfsrv_v4rootexport(void *argp, struct ucred *cred, struct thread *p)
3259 struct nfsex_args *nfsexargp = (struct nfsex_args *)argp;
3261 struct nameidata nd;
3264 error = vfs_export(&nfsv4root_mnt, &nfsexargp->export);
3265 if ((nfsexargp->export.ex_flags & MNT_DELEXPORT) != 0)
3267 else if (error == 0) {
3268 if (nfsexargp->fspec == NULL) {
3273 * If fspec != NULL, this is the v4root path.
3275 NDINIT(&nd, LOOKUP, FOLLOW, UIO_USERSPACE,
3276 nfsexargp->fspec, p);
3277 if ((error = namei(&nd)) != 0)
3279 error = nfsvno_getfh(nd.ni_vp, &fh, p);
3282 nfs_rootfh.nfsrvfh_len = NFSX_MYFH;
3283 NFSBCOPY((caddr_t)&fh,
3284 nfs_rootfh.nfsrvfh_data,
3285 sizeof (fhandle_t));
3296 * This function needs to test to see if the system is near its limit
3297 * for memory allocation via malloc() or mget() and return True iff
3298 * either of these resources are near their limit.
3299 * XXX (For now, this is just a stub.)
3301 int nfsrv_testmalloclimit = 0;
3303 nfsrv_mallocmget_limit(void)
3305 static int printmesg = 0;
3306 static int testval = 1;
3308 if (nfsrv_testmalloclimit && (testval++ % 1000) == 0) {
3309 if ((printmesg++ % 100) == 0)
3310 printf("nfsd: malloc/mget near limit\n");
3317 * BSD specific initialization of a mount point.
3322 static int inited = 0;
3327 nfsv4root_mnt.mnt_flag = (MNT_RDONLY | MNT_EXPORTED);
3328 TAILQ_INIT(&nfsv4root_mnt.mnt_nvnodelist);
3329 TAILQ_INIT(&nfsv4root_mnt.mnt_lazyvnodelist);
3330 nfsv4root_mnt.mnt_export = NULL;
3331 TAILQ_INIT(&nfsv4root_opt);
3332 TAILQ_INIT(&nfsv4root_newopt);
3333 nfsv4root_mnt.mnt_opt = &nfsv4root_opt;
3334 nfsv4root_mnt.mnt_optnew = &nfsv4root_newopt;
3335 nfsv4root_mnt.mnt_nvnodelistsize = 0;
3336 nfsv4root_mnt.mnt_lazyvnodelistsize = 0;
3340 * Get a vnode for a file handle, without checking exports, etc.
3343 nfsvno_getvp(fhandle_t *fhp)
3349 mp = vfs_busyfs(&fhp->fh_fsid);
3352 error = VFS_FHTOVP(mp, &fhp->fh_fid, LK_EXCLUSIVE, &vp);
3360 * Do a local VOP_ADVLOCK().
3363 nfsvno_advlock(struct vnode *vp, int ftype, u_int64_t first,
3364 u_int64_t end, struct thread *td)
3370 if (nfsrv_dolocallocks == 0)
3372 ASSERT_VOP_UNLOCKED(vp, "nfsvno_advlock: vp locked");
3374 fl.l_whence = SEEK_SET;
3376 fl.l_start = (off_t)first;
3377 if (end == NFS64BITSSET) {
3381 fl.l_len = (off_t)tlen;
3384 * For FreeBSD8, the l_pid and l_sysid must be set to the same
3385 * values for all calls, so that all locks will be held by the
3386 * nfsd server. (The nfsd server handles conflicts between the
3388 * Since an NFSv4 lockowner is a ClientID plus an array of up to 1024
3389 * bytes, so it can't be put in l_sysid.
3391 if (nfsv4_sysid == 0)
3392 nfsv4_sysid = nlm_acquire_next_sysid();
3393 fl.l_pid = (pid_t)0;
3394 fl.l_sysid = (int)nfsv4_sysid;
3396 if (ftype == F_UNLCK)
3397 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_UNLCK, &fl,
3398 (F_POSIX | F_REMOTE));
3400 error = VOP_ADVLOCK(vp, (caddr_t)td->td_proc, F_SETLK, &fl,
3401 (F_POSIX | F_REMOTE));
3409 * Check the nfsv4 root exports.
3412 nfsvno_v4rootexport(struct nfsrv_descript *nd)
3414 struct ucred *credanon;
3415 int error = 0, numsecflavor, secflavors[MAXSECFLAVORS], i;
3418 error = vfs_stdcheckexp(&nfsv4root_mnt, nd->nd_nam, &exflags,
3419 &credanon, &numsecflavor, secflavors);
3421 error = NFSERR_PROGUNAVAIL;
3424 if (credanon != NULL)
3426 for (i = 0; i < numsecflavor; i++) {
3427 if (secflavors[i] == AUTH_SYS)
3428 nd->nd_flag |= ND_EXAUTHSYS;
3429 else if (secflavors[i] == RPCSEC_GSS_KRB5)
3430 nd->nd_flag |= ND_EXGSS;
3431 else if (secflavors[i] == RPCSEC_GSS_KRB5I)
3432 nd->nd_flag |= ND_EXGSSINTEGRITY;
3433 else if (secflavors[i] == RPCSEC_GSS_KRB5P)
3434 nd->nd_flag |= ND_EXGSSPRIVACY;
3443 * Nfs server pseudo system call for the nfsd's
3449 nfssvc_nfsd(struct thread *td, struct nfssvc_args *uap)
3452 struct nfsd_addsock_args sockarg;
3453 struct nfsd_nfsd_args nfsdarg;
3454 struct nfsd_nfsd_oargs onfsdarg;
3455 struct nfsd_pnfsd_args pnfsdarg;
3456 struct vnode *vp, *nvp, *curdvp;
3457 struct pnfsdsfile *pf;
3458 struct nfsdevice *ds, *fds;
3459 cap_rights_t rights;
3460 int buflen, error, ret;
3461 char *buf, *cp, *cp2, *cp3;
3462 char fname[PNFS_FILENAME_LEN + 1];
3464 if (uap->flag & NFSSVC_NFSDADDSOCK) {
3465 error = copyin(uap->argp, (caddr_t)&sockarg, sizeof (sockarg));
3469 * Since we don't know what rights might be required,
3470 * pretend that we need them all. It is better to be too
3471 * careful than too reckless.
3473 error = fget(td, sockarg.sock,
3474 cap_rights_init(&rights, CAP_SOCK_SERVER), &fp);
3477 if (fp->f_type != DTYPE_SOCKET) {
3482 error = nfsrvd_addsock(fp);
3484 } else if (uap->flag & NFSSVC_NFSDNFSD) {
3485 if (uap->argp == NULL) {
3489 if ((uap->flag & NFSSVC_NEWSTRUCT) == 0) {
3490 error = copyin(uap->argp, &onfsdarg, sizeof(onfsdarg));
3492 nfsdarg.principal = onfsdarg.principal;
3493 nfsdarg.minthreads = onfsdarg.minthreads;
3494 nfsdarg.maxthreads = onfsdarg.maxthreads;
3495 nfsdarg.version = 1;
3496 nfsdarg.addr = NULL;
3497 nfsdarg.addrlen = 0;
3498 nfsdarg.dnshost = NULL;
3499 nfsdarg.dnshostlen = 0;
3500 nfsdarg.dspath = NULL;
3501 nfsdarg.dspathlen = 0;
3502 nfsdarg.mdspath = NULL;
3503 nfsdarg.mdspathlen = 0;
3504 nfsdarg.mirrorcnt = 1;
3507 error = copyin(uap->argp, &nfsdarg, sizeof(nfsdarg));
3510 if (nfsdarg.addrlen > 0 && nfsdarg.addrlen < 10000 &&
3511 nfsdarg.dnshostlen > 0 && nfsdarg.dnshostlen < 10000 &&
3512 nfsdarg.dspathlen > 0 && nfsdarg.dspathlen < 10000 &&
3513 nfsdarg.mdspathlen > 0 && nfsdarg.mdspathlen < 10000 &&
3514 nfsdarg.mirrorcnt >= 1 &&
3515 nfsdarg.mirrorcnt <= NFSDEV_MAXMIRRORS &&
3516 nfsdarg.addr != NULL && nfsdarg.dnshost != NULL &&
3517 nfsdarg.dspath != NULL && nfsdarg.mdspath != NULL) {
3518 NFSD_DEBUG(1, "addrlen=%d dspathlen=%d dnslen=%d"
3519 " mdspathlen=%d mirrorcnt=%d\n", nfsdarg.addrlen,
3520 nfsdarg.dspathlen, nfsdarg.dnshostlen,
3521 nfsdarg.mdspathlen, nfsdarg.mirrorcnt);
3522 cp = malloc(nfsdarg.addrlen + 1, M_TEMP, M_WAITOK);
3523 error = copyin(nfsdarg.addr, cp, nfsdarg.addrlen);
3528 cp[nfsdarg.addrlen] = '\0'; /* Ensure nul term. */
3530 cp = malloc(nfsdarg.dnshostlen + 1, M_TEMP, M_WAITOK);
3531 error = copyin(nfsdarg.dnshost, cp, nfsdarg.dnshostlen);
3533 free(nfsdarg.addr, M_TEMP);
3537 cp[nfsdarg.dnshostlen] = '\0'; /* Ensure nul term. */
3538 nfsdarg.dnshost = cp;
3539 cp = malloc(nfsdarg.dspathlen + 1, M_TEMP, M_WAITOK);
3540 error = copyin(nfsdarg.dspath, cp, nfsdarg.dspathlen);
3542 free(nfsdarg.addr, M_TEMP);
3543 free(nfsdarg.dnshost, M_TEMP);
3547 cp[nfsdarg.dspathlen] = '\0'; /* Ensure nul term. */
3548 nfsdarg.dspath = cp;
3549 cp = malloc(nfsdarg.mdspathlen + 1, M_TEMP, M_WAITOK);
3550 error = copyin(nfsdarg.mdspath, cp, nfsdarg.mdspathlen);
3552 free(nfsdarg.addr, M_TEMP);
3553 free(nfsdarg.dnshost, M_TEMP);
3554 free(nfsdarg.dspath, M_TEMP);
3558 cp[nfsdarg.mdspathlen] = '\0'; /* Ensure nul term. */
3559 nfsdarg.mdspath = cp;
3561 nfsdarg.addr = NULL;
3562 nfsdarg.addrlen = 0;
3563 nfsdarg.dnshost = NULL;
3564 nfsdarg.dnshostlen = 0;
3565 nfsdarg.dspath = NULL;
3566 nfsdarg.dspathlen = 0;
3567 nfsdarg.mdspath = NULL;
3568 nfsdarg.mdspathlen = 0;
3569 nfsdarg.mirrorcnt = 1;
3571 error = nfsrvd_nfsd(td, &nfsdarg);
3572 free(nfsdarg.addr, M_TEMP);
3573 free(nfsdarg.dnshost, M_TEMP);
3574 free(nfsdarg.dspath, M_TEMP);
3575 free(nfsdarg.mdspath, M_TEMP);
3576 } else if (uap->flag & NFSSVC_PNFSDS) {
3577 error = copyin(uap->argp, &pnfsdarg, sizeof(pnfsdarg));
3578 if (error == 0 && (pnfsdarg.op == PNFSDOP_DELDSSERVER ||
3579 pnfsdarg.op == PNFSDOP_FORCEDELDS)) {
3580 cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3581 error = copyinstr(pnfsdarg.dspath, cp, PATH_MAX + 1,
3584 error = nfsrv_deldsserver(pnfsdarg.op, cp, td);
3586 } else if (error == 0 && pnfsdarg.op == PNFSDOP_COPYMR) {
3587 cp = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3588 buflen = sizeof(*pf) * NFSDEV_MAXMIRRORS;
3589 buf = malloc(buflen, M_TEMP, M_WAITOK);
3590 error = copyinstr(pnfsdarg.mdspath, cp, PATH_MAX + 1,
3592 NFSD_DEBUG(4, "pnfsdcopymr cp mdspath=%d\n", error);
3593 if (error == 0 && pnfsdarg.dspath != NULL) {
3594 cp2 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3595 error = copyinstr(pnfsdarg.dspath, cp2,
3596 PATH_MAX + 1, NULL);
3597 NFSD_DEBUG(4, "pnfsdcopymr cp dspath=%d\n",
3601 if (error == 0 && pnfsdarg.curdspath != NULL) {
3602 cp3 = malloc(PATH_MAX + 1, M_TEMP, M_WAITOK);
3603 error = copyinstr(pnfsdarg.curdspath, cp3,
3604 PATH_MAX + 1, NULL);
3605 NFSD_DEBUG(4, "pnfsdcopymr cp curdspath=%d\n",
3612 error = nfsrv_mdscopymr(cp, cp2, cp3, buf,
3613 &buflen, fname, td, &vp, &nvp, &pf, &ds,
3615 NFSD_DEBUG(4, "nfsrv_mdscopymr=%d\n", error);
3617 if (pf->dsf_dir >= nfsrv_dsdirsize) {
3618 printf("copymr: dsdir out of range\n");
3621 NFSD_DEBUG(4, "copymr: buflen=%d\n", buflen);
3622 error = nfsrv_copymr(vp, nvp,
3623 ds->nfsdev_dsdir[pf->dsf_dir], ds, pf,
3624 (struct pnfsdsfile *)buf,
3625 buflen / sizeof(*pf), td->td_ucred, td);
3628 if (fds != NULL && error == 0) {
3629 curdvp = fds->nfsdev_dsdir[pf->dsf_dir];
3630 ret = vn_lock(curdvp, LK_EXCLUSIVE);
3632 nfsrv_dsremove(curdvp, fname,
3634 NFSVOPUNLOCK(curdvp);
3637 NFSD_DEBUG(4, "nfsrv_copymr=%d\n", error);
3645 error = nfssvc_srvcall(td, uap, td->td_ucred);
3654 nfssvc_srvcall(struct thread *p, struct nfssvc_args *uap, struct ucred *cred)
3656 struct nfsex_args export;
3657 struct nfsex_oldargs oexp;
3658 struct file *fp = NULL;
3659 int stablefd, i, len;
3660 struct nfsd_clid adminrevoke;
3661 struct nfsd_dumplist dumplist;
3662 struct nfsd_dumpclients *dumpclients;
3663 struct nfsd_dumplocklist dumplocklist;
3664 struct nfsd_dumplocks *dumplocks;
3665 struct nameidata nd;
3667 int error = EINVAL, igotlock;
3670 static int suspend_nfsd = 0;
3672 if (uap->flag & NFSSVC_PUBLICFH) {
3673 NFSBZERO((caddr_t)&nfs_pubfh.nfsrvfh_data,
3674 sizeof (fhandle_t));
3675 error = copyin(uap->argp,
3676 &nfs_pubfh.nfsrvfh_data, sizeof (fhandle_t));
3679 } else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) ==
3680 (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) {
3681 error = copyin(uap->argp,(caddr_t)&export,
3682 sizeof (struct nfsex_args));
3685 if (export.export.ex_ngroups > NGROUPS_MAX ||
3686 export.export.ex_ngroups < 0)
3688 else if (export.export.ex_ngroups > 0) {
3689 grps = malloc(export.export.ex_ngroups *
3690 sizeof(gid_t), M_TEMP, M_WAITOK);
3691 error = copyin(export.export.ex_groups, grps,
3692 export.export.ex_ngroups * sizeof(gid_t));
3693 export.export.ex_groups = grps;
3695 export.export.ex_groups = NULL;
3697 error = nfsrv_v4rootexport(&export, cred, p);
3700 } else if ((uap->flag & (NFSSVC_V4ROOTEXPORT | NFSSVC_NEWSTRUCT)) ==
3701 NFSSVC_V4ROOTEXPORT) {
3702 error = copyin(uap->argp,(caddr_t)&oexp,
3703 sizeof (struct nfsex_oldargs));
3705 memset(&export.export, 0, sizeof(export.export));
3706 export.export.ex_flags = (uint64_t)oexp.export.ex_flags;
3707 export.export.ex_root = oexp.export.ex_root;
3708 export.export.ex_uid = oexp.export.ex_anon.cr_uid;
3709 export.export.ex_ngroups =
3710 oexp.export.ex_anon.cr_ngroups;
3711 export.export.ex_groups = NULL;
3712 if (export.export.ex_ngroups > XU_NGROUPS ||
3713 export.export.ex_ngroups < 0)
3715 else if (export.export.ex_ngroups > 0) {
3716 export.export.ex_groups = malloc(
3717 export.export.ex_ngroups * sizeof(gid_t),
3719 for (i = 0; i < export.export.ex_ngroups; i++)
3720 export.export.ex_groups[i] =
3721 oexp.export.ex_anon.cr_groups[i];
3723 export.export.ex_addr = oexp.export.ex_addr;
3724 export.export.ex_addrlen = oexp.export.ex_addrlen;
3725 export.export.ex_mask = oexp.export.ex_mask;
3726 export.export.ex_masklen = oexp.export.ex_masklen;
3727 export.export.ex_indexfile = oexp.export.ex_indexfile;
3728 export.export.ex_numsecflavors =
3729 oexp.export.ex_numsecflavors;
3730 if (export.export.ex_numsecflavors >= MAXSECFLAVORS ||
3731 export.export.ex_numsecflavors < 0)
3734 for (i = 0; i < export.export.ex_numsecflavors;
3736 export.export.ex_secflavors[i] =
3737 oexp.export.ex_secflavors[i];
3739 export.fspec = oexp.fspec;
3741 error = nfsrv_v4rootexport(&export, cred, p);
3742 free(export.export.ex_groups, M_TEMP);
3744 } else if (uap->flag & NFSSVC_NOPUBLICFH) {
3747 } else if (uap->flag & NFSSVC_STABLERESTART) {
3748 error = copyin(uap->argp, (caddr_t)&stablefd,
3751 error = fp_getfvp(p, stablefd, &fp, &vp);
3752 if (!error && (NFSFPFLAG(fp) & (FREAD | FWRITE)) != (FREAD | FWRITE))
3754 if (!error && newnfs_numnfsd != 0)
3757 nfsrv_stablefirst.nsf_fp = fp;
3758 nfsrv_setupstable(p);
3760 } else if (uap->flag & NFSSVC_ADMINREVOKE) {
3761 error = copyin(uap->argp, (caddr_t)&adminrevoke,
3762 sizeof (struct nfsd_clid));
3764 error = nfsrv_adminrevoke(&adminrevoke, p);
3765 } else if (uap->flag & NFSSVC_DUMPCLIENTS) {
3766 error = copyin(uap->argp, (caddr_t)&dumplist,
3767 sizeof (struct nfsd_dumplist));
3768 if (!error && (dumplist.ndl_size < 1 ||
3769 dumplist.ndl_size > NFSRV_MAXDUMPLIST))
3772 len = sizeof (struct nfsd_dumpclients) * dumplist.ndl_size;
3773 dumpclients = malloc(len, M_TEMP, M_WAITOK | M_ZERO);
3774 nfsrv_dumpclients(dumpclients, dumplist.ndl_size);
3775 error = copyout(dumpclients, dumplist.ndl_list, len);
3776 free(dumpclients, M_TEMP);
3778 } else if (uap->flag & NFSSVC_DUMPLOCKS) {
3779 error = copyin(uap->argp, (caddr_t)&dumplocklist,
3780 sizeof (struct nfsd_dumplocklist));
3781 if (!error && (dumplocklist.ndllck_size < 1 ||
3782 dumplocklist.ndllck_size > NFSRV_MAXDUMPLIST))
3785 error = nfsrv_lookupfilename(&nd,
3786 dumplocklist.ndllck_fname, p);
3788 len = sizeof (struct nfsd_dumplocks) *
3789 dumplocklist.ndllck_size;
3790 dumplocks = malloc(len, M_TEMP, M_WAITOK | M_ZERO);
3791 nfsrv_dumplocks(nd.ni_vp, dumplocks,
3792 dumplocklist.ndllck_size, p);
3794 error = copyout(dumplocks, dumplocklist.ndllck_list,
3796 free(dumplocks, M_TEMP);
3798 } else if (uap->flag & NFSSVC_BACKUPSTABLE) {
3801 nfsd_master_pid = procp->p_pid;
3802 bcopy(procp->p_comm, nfsd_master_comm, MAXCOMLEN + 1);
3803 nfsd_master_start = procp->p_stats->p_start;
3804 nfsd_master_proc = procp;
3806 } else if ((uap->flag & NFSSVC_SUSPENDNFSD) != 0) {
3807 NFSLOCKV4ROOTMUTEX();
3808 if (suspend_nfsd == 0) {
3809 /* Lock out all nfsd threads */
3811 igotlock = nfsv4_lock(&nfsd_suspend_lock, 1,
3812 NULL, NFSV4ROOTLOCKMUTEXPTR, NULL);
3813 } while (igotlock == 0 && suspend_nfsd == 0);
3816 NFSUNLOCKV4ROOTMUTEX();
3818 } else if ((uap->flag & NFSSVC_RESUMENFSD) != 0) {
3819 NFSLOCKV4ROOTMUTEX();
3820 if (suspend_nfsd != 0) {
3821 nfsv4_unlock(&nfsd_suspend_lock, 0);
3824 NFSUNLOCKV4ROOTMUTEX();
3834 * Returns 0 if ok, 1 otherwise.
3837 nfsvno_testexp(struct nfsrv_descript *nd, struct nfsexstuff *exp)
3842 * This seems odd, but allow the case where the security flavor
3843 * list is empty. This happens when NFSv4 is traversing non-exported
3844 * file systems. Exported file systems should always have a non-empty
3845 * security flavor list.
3847 if (exp->nes_numsecflavor == 0)
3850 for (i = 0; i < exp->nes_numsecflavor; i++) {
3852 * The tests for privacy and integrity must be first,
3853 * since ND_GSS is set for everything but AUTH_SYS.
3855 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5P &&
3856 (nd->nd_flag & ND_GSSPRIVACY))
3858 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5I &&
3859 (nd->nd_flag & ND_GSSINTEGRITY))
3861 if (exp->nes_secflavors[i] == RPCSEC_GSS_KRB5 &&
3862 (nd->nd_flag & ND_GSS))
3864 if (exp->nes_secflavors[i] == AUTH_SYS &&
3865 (nd->nd_flag & ND_GSS) == 0)
3872 * Calculate a hash value for the fid in a file handle.
3875 nfsrv_hashfh(fhandle_t *fhp)
3879 hashval = hash32_buf(&fhp->fh_fid, sizeof(struct fid), 0);
3884 * Calculate a hash value for the sessionid.
3887 nfsrv_hashsessionid(uint8_t *sessionid)
3891 hashval = hash32_buf(sessionid, NFSX_V4SESSIONID, 0);
3896 * Signal the userland master nfsd to backup the stable restart file.
3899 nfsrv_backupstable(void)
3903 if (nfsd_master_proc != NULL) {
3904 procp = pfind(nfsd_master_pid);
3905 /* Try to make sure it is the correct process. */
3906 if (procp == nfsd_master_proc &&
3907 procp->p_stats->p_start.tv_sec ==
3908 nfsd_master_start.tv_sec &&
3909 procp->p_stats->p_start.tv_usec ==
3910 nfsd_master_start.tv_usec &&
3911 strcmp(procp->p_comm, nfsd_master_comm) == 0)
3912 kern_psignal(procp, SIGUSR2);
3914 nfsd_master_proc = NULL;
3922 * Create a DS data file for nfsrv_pnfscreate(). Called for each mirror.
3923 * The arguments are in a structure, so that they can be passed through
3924 * taskqueue for a kernel process to execute this function.
3926 struct nfsrvdscreate {
3930 struct ucred *tcred;
3933 struct pnfsdsfile *pf;
3937 struct vattr createva;
3941 nfsrv_dscreate(struct vnode *dvp, struct vattr *vap, struct vattr *nvap,
3942 fhandle_t *fhp, struct pnfsdsfile *pf, struct pnfsdsattr *dsa,
3943 char *fnamep, struct ucred *tcred, NFSPROC_T *p, struct vnode **nvpp)
3946 struct nameidata named;
3951 struct nfsmount *nmp;
3954 NFSNAMEICNDSET(&named.ni_cnd, tcred, CREATE,
3955 LOCKPARENT | LOCKLEAF | SAVESTART | NOCACHE);
3956 nfsvno_setpathbuf(&named, &bufp, &hashp);
3957 named.ni_cnd.cn_lkflags = LK_EXCLUSIVE;
3958 named.ni_cnd.cn_thread = p;
3959 named.ni_cnd.cn_nameptr = bufp;
3960 if (fnamep != NULL) {
3961 strlcpy(bufp, fnamep, PNFS_FILENAME_LEN + 1);
3962 named.ni_cnd.cn_namelen = strlen(bufp);
3964 named.ni_cnd.cn_namelen = nfsrv_putfhname(fhp, bufp);
3965 NFSD_DEBUG(4, "nfsrv_dscreate: dvp=%p fname=%s\n", dvp, bufp);
3967 /* Create the date file in the DS mount. */
3968 error = NFSVOPLOCK(dvp, LK_EXCLUSIVE);
3970 error = VOP_CREATE(dvp, &nvp, &named.ni_cnd, vap);
3973 /* Set the ownership of the file. */
3974 error = VOP_SETATTR(nvp, nvap, tcred);
3975 NFSD_DEBUG(4, "nfsrv_dscreate:"
3976 " setattr-uid=%d\n", error);
3981 printf("pNFS: pnfscreate failed=%d\n", error);
3983 printf("pNFS: pnfscreate vnlock=%d\n", error);
3986 nmp = VFSTONFS(nvp->v_mount);
3987 if (strcmp(nvp->v_mount->mnt_vfc->vfc_name, "nfs")
3988 != 0 || nmp->nm_nam->sa_len > sizeof(
3989 struct sockaddr_in6) ||
3990 np->n_fhp->nfh_len != NFSX_MYFH) {
3991 printf("Bad DS file: fstype=%s salen=%d"
3993 nvp->v_mount->mnt_vfc->vfc_name,
3994 nmp->nm_nam->sa_len, np->n_fhp->nfh_len);
3998 /* Set extattrs for the DS on the MDS file. */
4001 error = VOP_GETATTR(nvp, &va, tcred);
4003 dsa->dsa_filerev = va.va_filerev;
4004 dsa->dsa_size = va.va_size;
4005 dsa->dsa_atime = va.va_atime;
4006 dsa->dsa_mtime = va.va_mtime;
4007 dsa->dsa_bytes = va.va_bytes;
4011 NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh,
4013 NFSBCOPY(nmp->nm_nam, &pf->dsf_sin,
4014 nmp->nm_nam->sa_len);
4015 NFSBCOPY(named.ni_cnd.cn_nameptr,
4017 sizeof(pf->dsf_filename));
4020 printf("pNFS: pnfscreate can't get DS"
4021 " attr=%d\n", error);
4022 if (nvpp != NULL && error == 0)
4027 nfsvno_relpathbuf(&named);
4032 * Start up the thread that will execute nfsrv_dscreate().
4035 start_dscreate(void *arg, int pending)
4037 struct nfsrvdscreate *dsc;
4039 dsc = (struct nfsrvdscreate *)arg;
4040 dsc->err = nfsrv_dscreate(dsc->dvp, &dsc->createva, &dsc->va, &dsc->fh,
4041 dsc->pf, NULL, NULL, dsc->tcred, dsc->p, NULL);
4043 NFSD_DEBUG(4, "start_dscreate: err=%d\n", dsc->err);
4047 * Create a pNFS data file on the Data Server(s).
4050 nfsrv_pnfscreate(struct vnode *vp, struct vattr *vap, struct ucred *cred,
4053 struct nfsrvdscreate *dsc, *tdsc = NULL;
4054 struct nfsdevice *ds, *tds, *fds;
4056 struct pnfsdsfile *pf, *tpf;
4057 struct pnfsdsattr dsattr;
4059 struct vnode *dvp[NFSDEV_MAXMIRRORS];
4060 struct nfsmount *nmp;
4065 struct ucred *tcred;
4066 int dsdir[NFSDEV_MAXMIRRORS], error, i, mirrorcnt, ret;
4069 /* Get a DS server directory in a round-robin order. */
4075 * Search for the first entry that handles this MDS fs, but use the
4076 * first entry for all MDS fs's otherwise.
4078 TAILQ_FOREACH(tds, &nfsrv_devidhead, nfsdev_list) {
4079 if (tds->nfsdev_nmp != NULL) {
4080 if (tds->nfsdev_mdsisset == 0 && ds == NULL)
4082 else if (tds->nfsdev_mdsisset != 0 && fsidcmp(
4083 &mp->mnt_stat.f_fsid, &tds->nfsdev_mdsfsid) == 0) {
4091 NFSD_DEBUG(4, "nfsrv_pnfscreate: no srv\n");
4094 i = dsdir[0] = ds->nfsdev_nextdir;
4095 ds->nfsdev_nextdir = (ds->nfsdev_nextdir + 1) % nfsrv_dsdirsize;
4096 dvp[0] = ds->nfsdev_dsdir[i];
4097 tds = TAILQ_NEXT(ds, nfsdev_list);
4098 if (nfsrv_maxpnfsmirror > 1 && tds != NULL) {
4099 TAILQ_FOREACH_FROM(tds, &nfsrv_devidhead, nfsdev_list) {
4100 if (tds->nfsdev_nmp != NULL &&
4101 ((tds->nfsdev_mdsisset == 0 && fds == NULL) ||
4102 (tds->nfsdev_mdsisset != 0 && fds != NULL &&
4103 fsidcmp(&mp->mnt_stat.f_fsid,
4104 &tds->nfsdev_mdsfsid) == 0))) {
4105 dsdir[mirrorcnt] = i;
4106 dvp[mirrorcnt] = tds->nfsdev_dsdir[i];
4108 if (mirrorcnt >= nfsrv_maxpnfsmirror)
4113 /* Put at end of list to implement round-robin usage. */
4114 TAILQ_REMOVE(&nfsrv_devidhead, ds, nfsdev_list);
4115 TAILQ_INSERT_TAIL(&nfsrv_devidhead, ds, nfsdev_list);
4119 tdsc = dsc = malloc(sizeof(*dsc) * (mirrorcnt - 1), M_TEMP,
4121 tpf = pf = malloc(sizeof(*pf) * nfsrv_maxpnfsmirror, M_TEMP, M_WAITOK |
4124 error = nfsvno_getfh(vp, &fh, p);
4126 error = VOP_GETATTR(vp, &va, cred);
4128 /* Set the attributes for "vp" to Setattr the DS vp. */
4131 vamode = va.va_mode;
4135 va.va_mode = vamode;
4138 printf("pNFS: pnfscreate getfh+attr=%d\n", error);
4140 NFSD_DEBUG(4, "nfsrv_pnfscreate: cruid=%d crgid=%d\n", cred->cr_uid,
4142 /* Make data file name based on FH. */
4143 tcred = newnfs_getcred();
4146 * Create the file on each DS mirror, using kernel process(es) for the
4147 * additional mirrors.
4150 for (i = 0; i < mirrorcnt - 1 && error == 0; i++, tpf++, tdsc++) {
4151 tpf->dsf_dir = dsdir[i];
4152 tdsc->tcred = tcred;
4155 tdsc->createva = *vap;
4156 NFSBCOPY(&fh, &tdsc->fh, sizeof(fh));
4163 if (nfs_pnfsiothreads != 0) {
4164 ret = nfs_pnfsio(start_dscreate, tdsc);
4165 NFSD_DEBUG(4, "nfsrv_pnfscreate: nfs_pnfsio=%d\n", ret);
4168 ret = nfsrv_dscreate(dvp[i], vap, &va, &fh, tpf, NULL,
4169 NULL, tcred, p, NULL);
4171 KASSERT(error == 0, ("nfsrv_dscreate err=%d",
4173 if (failpos == -1 && nfsds_failerr(ret))
4181 tpf->dsf_dir = dsdir[mirrorcnt - 1];
4182 error = nfsrv_dscreate(dvp[mirrorcnt - 1], vap, &va, &fh, tpf,
4183 &dsattr, NULL, tcred, p, NULL);
4184 if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(error)) {
4185 failpos = mirrorcnt - 1;
4189 timo = hz / 50; /* Wait for 20msec. */
4192 /* Wait for kernel task(s) to complete. */
4193 for (tdsc = dsc, i = 0; i < mirrorcnt - 1; i++, tdsc++) {
4194 while (tdsc->inprog != 0 && tdsc->done == 0)
4195 tsleep(&tdsc->tsk, PVFS, "srvdcr", timo);
4196 if (tdsc->err != 0) {
4197 if (failpos == -1 && nfsds_failerr(tdsc->err))
4199 else if (error == 0)
4205 * If failpos has been set, that mirror has failed, so it needs
4209 nmp = VFSTONFS(dvp[failpos]->v_mount);
4211 if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
4212 NFSMNTP_CANCELRPCS)) == 0) {
4213 nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
4215 ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p);
4216 NFSD_DEBUG(4, "dscreatfail fail=%d ds=%p\n", failpos,
4219 nfsrv_killrpcs(nmp);
4221 nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
4229 ASSERT_VOP_ELOCKED(vp, "nfsrv_pnfscreate vp");
4231 NFSD_DEBUG(4, "nfsrv_pnfscreate: mirrorcnt=%d maxmirror=%d\n",
4232 mirrorcnt, nfsrv_maxpnfsmirror);
4234 * For all mirrors that couldn't be created, fill in the
4235 * *pf structure, but with an IP address == 0.0.0.0.
4237 tpf = pf + mirrorcnt;
4238 for (i = mirrorcnt; i < nfsrv_maxpnfsmirror; i++, tpf++) {
4240 tpf->dsf_sin.sin_family = AF_INET;
4241 tpf->dsf_sin.sin_len = sizeof(struct sockaddr_in);
4242 tpf->dsf_sin.sin_addr.s_addr = 0;
4243 tpf->dsf_sin.sin_port = 0;
4246 error = vn_extattr_set(vp, IO_NODELOCKED,
4247 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile",
4248 sizeof(*pf) * nfsrv_maxpnfsmirror, (char *)pf, p);
4250 error = vn_extattr_set(vp, IO_NODELOCKED,
4251 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr",
4252 sizeof(dsattr), (char *)&dsattr, p);
4254 printf("pNFS: pnfscreate setextattr=%d\n",
4257 printf("pNFS: pnfscreate=%d\n", error);
4263 * Get the information needed to remove the pNFS Data Server file from the
4264 * Metadata file. Upon success, ddvp is set non-NULL to the locked
4265 * DS directory vnode. The caller must unlock *ddvp when done with it.
4268 nfsrv_pnfsremovesetup(struct vnode *vp, NFSPROC_T *p, struct vnode **dvpp,
4269 int *mirrorcntp, char *fname, fhandle_t *fhp)
4272 struct ucred *tcred;
4277 /* If not an exported regular file or not a pNFS server, just return. */
4278 if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 ||
4279 nfsrv_devidcnt == 0)
4282 /* Check to see if this is the last hard link. */
4283 tcred = newnfs_getcred();
4284 error = VOP_GETATTR(vp, &va, tcred);
4287 printf("pNFS: nfsrv_pnfsremovesetup getattr=%d\n", error);
4290 if (va.va_nlink > 1)
4293 error = nfsvno_getfh(vp, fhp, p);
4295 printf("pNFS: nfsrv_pnfsremovesetup getfh=%d\n", error);
4300 buf = malloc(buflen, M_TEMP, M_WAITOK);
4301 /* Get the directory vnode for the DS mount and the file handle. */
4302 error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, dvpp,
4303 NULL, NULL, fname, NULL, NULL, NULL, NULL, NULL);
4306 printf("pNFS: nfsrv_pnfsremovesetup getsockmnt=%d\n", error);
4310 * Remove a DS data file for nfsrv_pnfsremove(). Called for each mirror.
4311 * The arguments are in a structure, so that they can be passed through
4312 * taskqueue for a kernel process to execute this function.
4314 struct nfsrvdsremove {
4318 struct ucred *tcred;
4322 char fname[PNFS_FILENAME_LEN + 1];
4326 nfsrv_dsremove(struct vnode *dvp, char *fname, struct ucred *tcred,
4329 struct nameidata named;
4335 error = NFSVOPLOCK(dvp, LK_EXCLUSIVE);
4338 named.ni_cnd.cn_nameiop = DELETE;
4339 named.ni_cnd.cn_lkflags = LK_EXCLUSIVE | LK_RETRY;
4340 named.ni_cnd.cn_cred = tcred;
4341 named.ni_cnd.cn_thread = p;
4342 named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF | SAVENAME;
4343 nfsvno_setpathbuf(&named, &bufp, &hashp);
4344 named.ni_cnd.cn_nameptr = bufp;
4345 named.ni_cnd.cn_namelen = strlen(fname);
4346 strlcpy(bufp, fname, NAME_MAX);
4347 NFSD_DEBUG(4, "nfsrv_pnfsremove: filename=%s\n", bufp);
4348 error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd);
4349 NFSD_DEBUG(4, "nfsrv_pnfsremove: aft LOOKUP=%d\n", error);
4351 error = VOP_REMOVE(dvp, nvp, &named.ni_cnd);
4355 nfsvno_relpathbuf(&named);
4357 printf("pNFS: nfsrv_pnfsremove failed=%d\n", error);
4362 * Start up the thread that will execute nfsrv_dsremove().
4365 start_dsremove(void *arg, int pending)
4367 struct nfsrvdsremove *dsrm;
4369 dsrm = (struct nfsrvdsremove *)arg;
4370 dsrm->err = nfsrv_dsremove(dsrm->dvp, dsrm->fname, dsrm->tcred,
4373 NFSD_DEBUG(4, "start_dsremove: err=%d\n", dsrm->err);
4377 * Remove a pNFS data file from a Data Server.
4378 * nfsrv_pnfsremovesetup() must have been called before the MDS file was
4379 * removed to set up the dvp and fill in the FH.
4382 nfsrv_pnfsremove(struct vnode **dvp, int mirrorcnt, char *fname, fhandle_t *fhp,
4385 struct ucred *tcred;
4386 struct nfsrvdsremove *dsrm, *tdsrm;
4387 struct nfsdevice *ds;
4388 struct nfsmount *nmp;
4389 int failpos, i, ret, timo;
4391 tcred = newnfs_getcred();
4394 dsrm = malloc(sizeof(*dsrm) * mirrorcnt - 1, M_TEMP, M_WAITOK);
4396 * Remove the file on each DS mirror, using kernel process(es) for the
4397 * additional mirrors.
4400 for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) {
4401 tdsrm->tcred = tcred;
4403 tdsrm->dvp = dvp[i];
4404 strlcpy(tdsrm->fname, fname, PNFS_FILENAME_LEN + 1);
4409 if (nfs_pnfsiothreads != 0) {
4410 ret = nfs_pnfsio(start_dsremove, tdsrm);
4411 NFSD_DEBUG(4, "nfsrv_pnfsremove: nfs_pnfsio=%d\n", ret);
4414 ret = nfsrv_dsremove(dvp[i], fname, tcred, p);
4415 if (failpos == -1 && nfsds_failerr(ret))
4419 ret = nfsrv_dsremove(dvp[mirrorcnt - 1], fname, tcred, p);
4420 if (failpos == -1 && mirrorcnt > 1 && nfsds_failerr(ret))
4421 failpos = mirrorcnt - 1;
4422 timo = hz / 50; /* Wait for 20msec. */
4425 /* Wait for kernel task(s) to complete. */
4426 for (tdsrm = dsrm, i = 0; i < mirrorcnt - 1; i++, tdsrm++) {
4427 while (tdsrm->inprog != 0 && tdsrm->done == 0)
4428 tsleep(&tdsrm->tsk, PVFS, "srvdsrm", timo);
4429 if (failpos == -1 && nfsds_failerr(tdsrm->err))
4434 * If failpos has been set, that mirror has failed, so it needs
4438 nmp = VFSTONFS(dvp[failpos]->v_mount);
4440 if ((nmp->nm_privflag & (NFSMNTP_FORCEDISM |
4441 NFSMNTP_CANCELRPCS)) == 0) {
4442 nmp->nm_privflag |= NFSMNTP_CANCELRPCS;
4444 ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER, nmp, p);
4445 NFSD_DEBUG(4, "dsremovefail fail=%d ds=%p\n", failpos,
4448 nfsrv_killrpcs(nmp);
4450 nmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
4456 /* Get rid all layouts for the file. */
4457 nfsrv_freefilelayouts(fhp);
4464 * Generate a file name based on the file handle and put it in *bufp.
4465 * Return the number of bytes generated.
4468 nfsrv_putfhname(fhandle_t *fhp, char *bufp)
4472 const uint8_t *hexdigits = "0123456789abcdef";
4474 cp = (uint8_t *)fhp;
4475 for (i = 0; i < sizeof(*fhp); i++) {
4476 bufp[2 * i] = hexdigits[(*cp >> 4) & 0xf];
4477 bufp[2 * i + 1] = hexdigits[*cp++ & 0xf];
4484 * Update the Metadata file's attributes from the DS file when a Read/Write
4485 * layout is returned.
4486 * Basically just call nfsrv_proxyds() with procedure == NFSPROC_LAYOUTRETURN
4487 * so that it does a nfsrv_getattrdsrpc() and nfsrv_setextattr() on the DS file.
4490 nfsrv_updatemdsattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p)
4492 struct ucred *tcred;
4495 /* Do this as root so that it won't fail with EACCES. */
4496 tcred = newnfs_getcred();
4497 error = nfsrv_proxyds(vp, 0, 0, tcred, p, NFSPROC_LAYOUTRETURN,
4498 NULL, NULL, NULL, nap, NULL, NULL, 0, NULL);
4504 * Set the NFSv4 ACL on the DS file to the same ACL as the MDS file.
4507 nfsrv_dssetacl(struct vnode *vp, struct acl *aclp, struct ucred *cred,
4512 error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SETACL,
4513 NULL, NULL, NULL, NULL, aclp, NULL, 0, NULL);
4518 nfsrv_proxyds(struct vnode *vp, off_t off, int cnt, struct ucred *cred,
4519 struct thread *p, int ioproc, struct mbuf **mpp, char *cp,
4520 struct mbuf **mpp2, struct nfsvattr *nap, struct acl *aclp,
4521 off_t *offp, int content, bool *eofp)
4523 struct nfsmount *nmp[NFSDEV_MAXMIRRORS], *failnmp;
4524 fhandle_t fh[NFSDEV_MAXMIRRORS];
4525 struct vnode *dvp[NFSDEV_MAXMIRRORS];
4526 struct nfsdevice *ds;
4527 struct pnfsdsattr dsattr;
4528 struct opnfsdsattr odsattr;
4530 int buflen, error, failpos, i, mirrorcnt, origmircnt, trycnt;
4532 NFSD_DEBUG(4, "in nfsrv_proxyds\n");
4534 * If not a regular file, not exported or not a pNFS server,
4535 * just return ENOENT.
4537 if (vp->v_type != VREG || (vp->v_mount->mnt_flag & MNT_EXPORTED) == 0 ||
4538 nfsrv_devidcnt == 0)
4542 buf = malloc(buflen, M_TEMP, M_WAITOK);
4546 * For Getattr, get the Change attribute (va_filerev) and size (va_size)
4547 * from the MetaData file's extended attribute.
4549 if (ioproc == NFSPROC_GETATTR) {
4550 error = vn_extattr_get(vp, IO_NODELOCKED,
4551 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsattr", &buflen, buf,
4554 if (buflen == sizeof(odsattr)) {
4555 NFSBCOPY(buf, &odsattr, buflen);
4556 nap->na_filerev = odsattr.dsa_filerev;
4557 nap->na_size = odsattr.dsa_size;
4558 nap->na_atime = odsattr.dsa_atime;
4559 nap->na_mtime = odsattr.dsa_mtime;
4561 * Fake na_bytes by rounding up na_size.
4562 * Since we don't know the block size, just
4563 * use BLKDEV_IOSIZE.
4565 nap->na_bytes = (odsattr.dsa_size +
4566 BLKDEV_IOSIZE - 1) & ~(BLKDEV_IOSIZE - 1);
4567 } else if (buflen == sizeof(dsattr)) {
4568 NFSBCOPY(buf, &dsattr, buflen);
4569 nap->na_filerev = dsattr.dsa_filerev;
4570 nap->na_size = dsattr.dsa_size;
4571 nap->na_atime = dsattr.dsa_atime;
4572 nap->na_mtime = dsattr.dsa_mtime;
4573 nap->na_bytes = dsattr.dsa_bytes;
4579 * If nfsrv_pnfsgetdsattr is 0 or nfsrv_checkdsattr()
4580 * returns 0, just return now. nfsrv_checkdsattr()
4581 * returns 0 if there is no Read/Write layout
4582 * plus either an Open/Write_access or Write
4583 * delegation issued to a client for the file.
4585 if (nfsrv_pnfsgetdsattr == 0 ||
4586 nfsrv_checkdsattr(vp, p) == 0) {
4593 * Clear ENOATTR so the code below will attempt to do a
4594 * nfsrv_getattrdsrpc() to get the attributes and (re)create
4595 * the extended attribute.
4597 if (error == ENOATTR)
4606 if (ioproc == NFSPROC_READDS && NFSVOPISLOCKED(vp) ==
4608 printf("nfsrv_proxyds: Readds vp exclusively locked\n");
4609 error = nfsrv_dsgetsockmnt(vp, LK_SHARED, buf, &buflen,
4610 &mirrorcnt, p, dvp, fh, NULL, NULL, NULL, NULL, NULL,
4613 for (i = 0; i < mirrorcnt; i++)
4614 nmp[i] = VFSTONFS(dvp[i]->v_mount);
4616 printf("pNFS: proxy getextattr sockaddr=%d\n", error);
4618 printf("pNFS: nfsrv_dsgetsockmnt=%d\n", error);
4621 if (origmircnt == -1)
4622 origmircnt = mirrorcnt;
4624 * If failpos is set to a mirror#, then that mirror has
4625 * failed and will be disabled. For Read, Getattr and Seek, the
4626 * function only tries one mirror, so if that mirror has
4627 * failed, it will need to be retried. As such, increment
4628 * tryitagain for these cases.
4629 * For Write, Setattr and Setacl, the function tries all
4630 * mirrors and will not return an error for the case where
4631 * one mirror has failed. For these cases, the functioning
4632 * mirror(s) will have been modified, so a retry isn't
4633 * necessary. These functions will set failpos for the
4636 if (ioproc == NFSPROC_READDS) {
4637 error = nfsrv_readdsrpc(fh, off, cnt, cred, p, nmp[0],
4639 if (nfsds_failerr(error) && mirrorcnt > 1) {
4641 * Setting failpos will cause the mirror
4642 * to be disabled and then a retry of this
4649 } else if (ioproc == NFSPROC_WRITEDS)
4650 error = nfsrv_writedsrpc(fh, off, cnt, cred, p, vp,
4651 &nmp[0], mirrorcnt, mpp, cp, &failpos);
4652 else if (ioproc == NFSPROC_SETATTR)
4653 error = nfsrv_setattrdsrpc(fh, cred, p, vp, &nmp[0],
4654 mirrorcnt, nap, &failpos);
4655 else if (ioproc == NFSPROC_SETACL)
4656 error = nfsrv_setacldsrpc(fh, cred, p, vp, &nmp[0],
4657 mirrorcnt, aclp, &failpos);
4658 else if (ioproc == NFSPROC_SEEKDS) {
4659 error = nfsrv_seekdsrpc(fh, offp, content, eofp, cred,
4661 if (nfsds_failerr(error) && mirrorcnt > 1) {
4663 * Setting failpos will cause the mirror
4664 * to be disabled and then a retry of this
4671 } else if (ioproc == NFSPROC_ALLOCATE)
4672 error = nfsrv_allocatedsrpc(fh, off, *offp, cred, p, vp,
4673 &nmp[0], mirrorcnt, &failpos);
4675 error = nfsrv_getattrdsrpc(&fh[mirrorcnt - 1], cred, p,
4676 vp, nmp[mirrorcnt - 1], nap);
4677 if (nfsds_failerr(error) && mirrorcnt > 1) {
4679 * Setting failpos will cause the mirror
4680 * to be disabled and then a retry of this
4681 * getattr is required.
4683 failpos = mirrorcnt - 1;
4690 failnmp = nmp[failpos];
4691 NFSLOCKMNT(failnmp);
4692 if ((failnmp->nm_privflag & (NFSMNTP_FORCEDISM |
4693 NFSMNTP_CANCELRPCS)) == 0) {
4694 failnmp->nm_privflag |= NFSMNTP_CANCELRPCS;
4695 NFSUNLOCKMNT(failnmp);
4696 ds = nfsrv_deldsnmp(PNFSDOP_DELDSSERVER,
4698 NFSD_DEBUG(4, "dsldsnmp fail=%d ds=%p\n",
4701 nfsrv_killrpcs(failnmp);
4702 NFSLOCKMNT(failnmp);
4703 failnmp->nm_privflag &= ~NFSMNTP_CANCELRPCS;
4706 NFSUNLOCKMNT(failnmp);
4708 for (i = 0; i < mirrorcnt; i++)
4709 NFSVOPUNLOCK(dvp[i]);
4710 NFSD_DEBUG(4, "nfsrv_proxyds: aft RPC=%d trya=%d\n", error,
4712 /* Try the Read/Getattr again if a mirror was deleted. */
4713 if (ds != NULL && trycnt > 0 && trycnt < origmircnt)
4716 /* Return ENOENT for any Extended Attribute error. */
4720 NFSD_DEBUG(4, "nfsrv_proxyds: error=%d\n", error);
4725 * Get the DS mount point, fh and directory from the "pnfsd.dsfile" extended
4727 * newnmpp - If it points to a non-NULL nmp, that is the destination and needs
4728 * to be checked. If it points to a NULL nmp, then it returns
4729 * a suitable destination.
4730 * curnmp - If non-NULL, it is the source mount for the copy.
4733 nfsrv_dsgetsockmnt(struct vnode *vp, int lktype, char *buf, int *buflenp,
4734 int *mirrorcntp, NFSPROC_T *p, struct vnode **dvpp, fhandle_t *fhp,
4735 char *devid, char *fnamep, struct vnode **nvpp, struct nfsmount **newnmpp,
4736 struct nfsmount *curnmp, int *ippos, int *dsdirp)
4738 struct vnode *dvp, *nvp = NULL, **tdvpp;
4740 struct nfsmount *nmp, *newnmp;
4741 struct sockaddr *sad;
4742 struct sockaddr_in *sin;
4743 struct nfsdevice *ds, *tds, *fndds;
4744 struct pnfsdsfile *pf;
4746 int error, fhiszero, fnd, gotone, i, mirrorcnt;
4748 ASSERT_VOP_LOCKED(vp, "nfsrv_dsgetsockmnt vp");
4757 if (newnmpp != NULL)
4762 error = vn_extattr_get(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM,
4763 "pnfsd.dsfile", buflenp, buf, p);
4764 mirrorcnt = *buflenp / sizeof(*pf);
4765 if (error == 0 && (mirrorcnt < 1 || mirrorcnt > NFSDEV_MAXMIRRORS ||
4766 *buflenp != sizeof(*pf) * mirrorcnt))
4769 pf = (struct pnfsdsfile *)buf;
4770 /* If curnmp != NULL, check for a match in the mirror list. */
4771 if (curnmp != NULL && error == 0) {
4773 for (i = 0; i < mirrorcnt; i++, pf++) {
4774 sad = (struct sockaddr *)&pf->dsf_sin;
4775 if (nfsaddr2_match(sad, curnmp->nm_nam)) {
4787 pf = (struct pnfsdsfile *)buf;
4788 NFSD_DEBUG(4, "nfsrv_dsgetsockmnt: mirrorcnt=%d err=%d\n", mirrorcnt,
4790 for (i = 0; i < mirrorcnt && error == 0; i++, pf++) {
4792 sad = (struct sockaddr *)&pf->dsf_sin;
4794 dsdir = pf->dsf_dir;
4795 if (dsdir >= nfsrv_dsdirsize) {
4796 printf("nfsrv_dsgetsockmnt: dsdir=%d\n", dsdir);
4798 } else if (nvpp != NULL && newnmp != NULL &&
4799 nfsaddr2_match(sad, newnmp->nm_nam))
4802 if (ippos != NULL && curnmp == NULL &&
4803 sad->sa_family == AF_INET &&
4804 sin->sin_addr.s_addr == 0)
4806 if (NFSBCMP(&zerofh, &pf->dsf_fh, sizeof(zerofh)) == 0)
4808 /* Use the socket address to find the mount point. */
4811 /* Find a match for the IP address. */
4812 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
4813 if (ds->nfsdev_nmp != NULL) {
4814 dvp = ds->nfsdev_dvp;
4815 nmp = VFSTONFS(dvp->v_mount);
4816 if (nmp != ds->nfsdev_nmp)
4817 printf("different2 nmp %p %p\n",
4818 nmp, ds->nfsdev_nmp);
4819 if (nfsaddr2_match(sad, nmp->nm_nam)) {
4825 if (fndds != NULL && newnmpp != NULL &&
4827 /* Search for a place to make a mirror copy. */
4828 TAILQ_FOREACH(tds, &nfsrv_devidhead,
4830 if (tds->nfsdev_nmp != NULL &&
4832 ((tds->nfsdev_mdsisset == 0 &&
4833 fndds->nfsdev_mdsisset == 0) ||
4834 (tds->nfsdev_mdsisset != 0 &&
4835 fndds->nfsdev_mdsisset != 0 &&
4836 fsidcmp(&tds->nfsdev_mdsfsid,
4837 &mp->mnt_stat.f_fsid) == 0))) {
4838 *newnmpp = tds->nfsdev_nmp;
4844 * Move this entry to the end of the
4845 * list, so it won't be selected as
4846 * easily the next time.
4848 TAILQ_REMOVE(&nfsrv_devidhead, tds,
4850 TAILQ_INSERT_TAIL(&nfsrv_devidhead, tds,
4855 if (fndds != NULL) {
4856 dvp = fndds->nfsdev_dsdir[dsdir];
4857 if (lktype != 0 || fhiszero != 0 ||
4858 (nvpp != NULL && *nvpp == NULL)) {
4860 error = vn_lock(dvp,
4862 else if (lktype != 0)
4863 error = vn_lock(dvp, lktype);
4865 error = vn_lock(dvp, LK_SHARED);
4867 * If the file handle is all 0's, try to
4868 * do a Lookup against the DS to acquire
4870 * If dvpp == NULL or the Lookup fails,
4871 * unlock dvp after the call.
4873 if (error == 0 && (fhiszero != 0 ||
4874 (nvpp != NULL && *nvpp == NULL))) {
4875 error = nfsrv_pnfslookupds(vp,
4891 if (error != 0 || lktype == 0)
4897 NFSD_DEBUG(4, "gotone=%d\n", gotone);
4898 if (devid != NULL) {
4899 NFSBCOPY(fndds->nfsdev_deviceid,
4900 devid, NFSX_V4DEVICEID);
4901 devid += NFSX_V4DEVICEID;
4906 NFSBCOPY(&pf->dsf_fh, fhp++,
4908 if (fnamep != NULL && gotone == 1)
4911 sizeof(pf->dsf_filename));
4913 NFSD_DEBUG(4, "nfsrv_dsgetsockmnt "
4918 if (error == 0 && gotone == 0)
4921 NFSD_DEBUG(4, "eo nfsrv_dsgetsockmnt: gotone=%d err=%d\n", gotone,
4924 *mirrorcntp = gotone;
4926 if (gotone > 0 && dvpp != NULL) {
4928 * If the error didn't occur on the first one and
4929 * dvpp != NULL, the one(s) prior to the failure will
4930 * have locked dvp's that need to be unlocked.
4932 for (i = 0; i < gotone; i++) {
4933 NFSVOPUNLOCK(*dvpp);
4938 * If it found the vnode to be copied from before a failure,
4939 * it needs to be vput()'d.
4941 if (nvpp != NULL && *nvpp != NULL) {
4950 * Set the extended attribute for the Change attribute.
4953 nfsrv_setextattr(struct vnode *vp, struct nfsvattr *nap, NFSPROC_T *p)
4955 struct pnfsdsattr dsattr;
4958 ASSERT_VOP_ELOCKED(vp, "nfsrv_setextattr vp");
4959 dsattr.dsa_filerev = nap->na_filerev;
4960 dsattr.dsa_size = nap->na_size;
4961 dsattr.dsa_atime = nap->na_atime;
4962 dsattr.dsa_mtime = nap->na_mtime;
4963 dsattr.dsa_bytes = nap->na_bytes;
4964 error = vn_extattr_set(vp, IO_NODELOCKED, EXTATTR_NAMESPACE_SYSTEM,
4965 "pnfsd.dsattr", sizeof(dsattr), (char *)&dsattr, p);
4967 printf("pNFS: setextattr=%d\n", error);
4972 nfsrv_readdsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred,
4973 NFSPROC_T *p, struct nfsmount *nmp, struct mbuf **mpp, struct mbuf **mpendp)
4976 struct nfsrv_descript *nd;
4978 struct mbuf *m, *m2;
4979 int error = 0, retlen, tlen, trimlen;
4981 NFSD_DEBUG(4, "in nfsrv_readdsrpc\n");
4982 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
4985 * Use a stateid where other is an alternating 01010 pattern and
4986 * seqid is 0xffffffff. This value is not defined as special by
4987 * the RFC and is used by the FreeBSD NFS server to indicate an
4988 * MDS->DS proxy operation.
4990 st.other[0] = 0x55555555;
4991 st.other[1] = 0x55555555;
4992 st.other[2] = 0x55555555;
4993 st.seqid = 0xffffffff;
4994 nfscl_reqstart(nd, NFSPROC_READDS, nmp, (u_int8_t *)fhp, sizeof(*fhp),
4995 NULL, NULL, 0, 0, false);
4996 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
4997 NFSM_BUILD(tl, uint32_t *, NFSX_UNSIGNED * 3);
4998 txdr_hyper(off, tl);
4999 *(tl + 2) = txdr_unsigned(len);
5000 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
5001 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5006 if (nd->nd_repstat == 0) {
5007 NFSM_DISSECT(tl, u_int32_t *, NFSX_UNSIGNED);
5008 NFSM_STRSIZ(retlen, len);
5010 /* Trim off the pre-data XDR from the mbuf chain. */
5012 while (m != NULL && m != nd->nd_md) {
5013 if (m->m_next == nd->nd_md) {
5015 m_freem(nd->nd_mrep);
5016 nd->nd_mrep = m = nd->nd_md;
5021 printf("nfsrv_readdsrpc: busted mbuf list\n");
5027 * Now, adjust first mbuf so that any XDR before the
5028 * read data is skipped over.
5030 trimlen = nd->nd_dpos - mtod(m, char *);
5032 m->m_len -= trimlen;
5033 NFSM_DATAP(m, trimlen);
5037 * Truncate the mbuf chain at retlen bytes of data,
5038 * plus XDR padding that brings the length up to a
5041 tlen = NFSM_RNDUP(retlen);
5043 if (m->m_len >= tlen) {
5053 } while (m != NULL);
5055 printf("nfsrv_readdsrpc: busted mbuf list\n");
5064 error = nd->nd_repstat;
5066 /* If nd->nd_mrep is already NULL, this is a no-op. */
5067 m_freem(nd->nd_mrep);
5069 NFSD_DEBUG(4, "nfsrv_readdsrpc error=%d\n", error);
5074 * Do a write RPC on a DS data file, using this structure for the arguments,
5075 * so that this function can be executed by a separate kernel process.
5077 struct nfsrvwritedsdorpc {
5084 struct nfsmount *nmp;
5092 nfsrv_writedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off, int len,
5093 struct nfsvattr *nap, struct mbuf *m, struct ucred *cred, NFSPROC_T *p)
5096 struct nfsrv_descript *nd;
5097 nfsattrbit_t attrbits;
5099 int commit, error, retlen;
5101 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5102 nfscl_reqstart(nd, NFSPROC_WRITE, nmp, (u_int8_t *)fhp,
5103 sizeof(fhandle_t), NULL, NULL, 0, 0, false);
5106 * Use a stateid where other is an alternating 01010 pattern and
5107 * seqid is 0xffffffff. This value is not defined as special by
5108 * the RFC and is used by the FreeBSD NFS server to indicate an
5109 * MDS->DS proxy operation.
5111 st.other[0] = 0x55555555;
5112 st.other[1] = 0x55555555;
5113 st.other[2] = 0x55555555;
5114 st.seqid = 0xffffffff;
5115 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5116 NFSM_BUILD(tl, u_int32_t *, NFSX_HYPER + 2 * NFSX_UNSIGNED);
5117 txdr_hyper(off, tl);
5120 * Do all writes FileSync, since the server doesn't hold onto dirty
5121 * buffers. Since clients should be accessing the DS servers directly
5122 * using the pNFS layouts, this just needs to work correctly as a
5125 *tl++ = txdr_unsigned(NFSWRITE_FILESYNC);
5126 *tl = txdr_unsigned(len);
5127 NFSD_DEBUG(4, "nfsrv_writedsdorpc: len=%d\n", len);
5129 /* Put data in mbuf chain. */
5130 nd->nd_mb->m_next = m;
5132 /* Set nd_mb and nd_bpos to end of data. */
5133 while (m->m_next != NULL)
5136 nfsm_set(nd, m->m_len);
5137 NFSD_DEBUG(4, "nfsrv_writedsdorpc: lastmb len=%d\n", m->m_len);
5139 /* Do a Getattr for the attributes that change upon writing. */
5140 NFSZERO_ATTRBIT(&attrbits);
5141 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
5142 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
5143 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
5144 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
5145 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
5146 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
5147 *tl = txdr_unsigned(NFSV4OP_GETATTR);
5148 (void) nfsrv_putattrbit(nd, &attrbits);
5149 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p,
5150 cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5155 NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft writerpc=%d\n", nd->nd_repstat);
5156 /* Get rid of weak cache consistency data for now. */
5157 if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) ==
5158 (ND_NFSV4 | ND_V4WCCATTR)) {
5159 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5160 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5161 NFSD_DEBUG(4, "nfsrv_writedsdorpc: wcc attr=%d\n", error);
5165 * Get rid of Op# and status for next op.
5167 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5169 nd->nd_flag |= ND_NOMOREDATA;
5171 if (nd->nd_repstat == 0) {
5172 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED + NFSX_VERF);
5173 retlen = fxdr_unsigned(int, *tl++);
5174 commit = fxdr_unsigned(int, *tl);
5175 if (commit != NFSWRITE_FILESYNC)
5177 NFSD_DEBUG(4, "nfsrv_writedsdorpc:retlen=%d commit=%d err=%d\n",
5178 retlen, commit, error);
5180 error = nd->nd_repstat;
5181 /* We have no use for the Write Verifier since we use FileSync. */
5184 * Get the Change, Size, Access Time and Modify Time attributes and set
5185 * on the Metadata file, so its attributes will be what the file's
5186 * would be if it had been written.
5189 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5190 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5191 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5193 NFSD_DEBUG(4, "nfsrv_writedsdorpc: aft loadattr=%d\n", error);
5195 m_freem(nd->nd_mrep);
5197 NFSD_DEBUG(4, "nfsrv_writedsdorpc error=%d\n", error);
5202 * Start up the thread that will execute nfsrv_writedsdorpc().
5205 start_writedsdorpc(void *arg, int pending)
5207 struct nfsrvwritedsdorpc *drpc;
5209 drpc = (struct nfsrvwritedsdorpc *)arg;
5210 drpc->err = nfsrv_writedsdorpc(drpc->nmp, &drpc->fh, drpc->off,
5211 drpc->len, NULL, drpc->m, drpc->cred, drpc->p);
5213 NFSD_DEBUG(4, "start_writedsdorpc: err=%d\n", drpc->err);
5217 nfsrv_writedsrpc(fhandle_t *fhp, off_t off, int len, struct ucred *cred,
5218 NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5219 struct mbuf **mpp, char *cp, int *failposp)
5221 struct nfsrvwritedsdorpc *drpc, *tdrpc = NULL;
5224 int error, i, offs, ret, timo;
5226 NFSD_DEBUG(4, "in nfsrv_writedsrpc\n");
5227 KASSERT(*mpp != NULL, ("nfsrv_writedsrpc: NULL mbuf chain"));
5230 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5233 /* Calculate offset in mbuf chain that data starts. */
5234 offs = cp - mtod(*mpp, char *);
5235 NFSD_DEBUG(4, "nfsrv_writedsrpc: mcopy offs=%d len=%d\n", offs, len);
5238 * Do the write RPC for every DS, using a separate kernel process
5239 * for every DS except the last one.
5242 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5244 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5252 tdrpc->m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK);
5254 if (nfs_pnfsiothreads != 0) {
5255 ret = nfs_pnfsio(start_writedsdorpc, tdrpc);
5256 NFSD_DEBUG(4, "nfsrv_writedsrpc: nfs_pnfsio=%d\n",
5260 ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, NULL,
5262 if (nfsds_failerr(ret) && *failposp == -1)
5264 else if (error == 0 && ret != 0)
5270 m = m_copym(*mpp, offs, NFSM_RNDUP(len), M_WAITOK);
5271 ret = nfsrv_writedsdorpc(*nmpp, fhp, off, len, &na, m, cred, p);
5272 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5273 *failposp = mirrorcnt - 1;
5274 else if (error == 0 && ret != 0)
5277 error = nfsrv_setextattr(vp, &na, p);
5278 NFSD_DEBUG(4, "nfsrv_writedsrpc: aft setextat=%d\n", error);
5280 timo = hz / 50; /* Wait for 20msec. */
5283 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5284 /* Wait for RPCs on separate threads to complete. */
5285 while (tdrpc->inprog != 0 && tdrpc->done == 0)
5286 tsleep(&tdrpc->tsk, PVFS, "srvwrds", timo);
5287 if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5289 else if (error == 0 && tdrpc->err != 0)
5297 * Do a allocate RPC on a DS data file, using this structure for the arguments,
5298 * so that this function can be executed by a separate kernel process.
5300 struct nfsrvallocatedsdorpc {
5307 struct nfsmount *nmp;
5314 nfsrv_allocatedsdorpc(struct nfsmount *nmp, fhandle_t *fhp, off_t off,
5315 off_t len, struct nfsvattr *nap, struct ucred *cred, NFSPROC_T *p)
5318 struct nfsrv_descript *nd;
5319 nfsattrbit_t attrbits;
5323 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5324 nfscl_reqstart(nd, NFSPROC_ALLOCATE, nmp, (u_int8_t *)fhp,
5325 sizeof(fhandle_t), NULL, NULL, 0, 0, false);
5328 * Use a stateid where other is an alternating 01010 pattern and
5329 * seqid is 0xffffffff. This value is not defined as special by
5330 * the RFC and is used by the FreeBSD NFS server to indicate an
5331 * MDS->DS proxy operation.
5333 st.other[0] = 0x55555555;
5334 st.other[1] = 0x55555555;
5335 st.other[2] = 0x55555555;
5336 st.seqid = 0xffffffff;
5337 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5338 NFSM_BUILD(tl, uint32_t *, 2 * NFSX_HYPER + NFSX_UNSIGNED);
5339 txdr_hyper(off, tl); tl += 2;
5340 txdr_hyper(len, tl); tl += 2;
5341 NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: len=%jd\n", (intmax_t)len);
5343 *tl = txdr_unsigned(NFSV4OP_GETATTR);
5344 NFSGETATTR_ATTRBIT(&attrbits);
5345 nfsrv_putattrbit(nd, &attrbits);
5346 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p,
5347 cred, NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5352 NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft allocaterpc=%d\n",
5354 if (nd->nd_repstat == 0) {
5355 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5356 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0, NULL, NULL,
5357 NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5359 error = nd->nd_repstat;
5360 NFSD_DEBUG(4, "nfsrv_allocatedsdorpc: aft loadattr=%d\n", error);
5362 m_freem(nd->nd_mrep);
5364 NFSD_DEBUG(4, "nfsrv_allocatedsdorpc error=%d\n", error);
5369 * Start up the thread that will execute nfsrv_allocatedsdorpc().
5372 start_allocatedsdorpc(void *arg, int pending)
5374 struct nfsrvallocatedsdorpc *drpc;
5376 drpc = (struct nfsrvallocatedsdorpc *)arg;
5377 drpc->err = nfsrv_allocatedsdorpc(drpc->nmp, &drpc->fh, drpc->off,
5378 drpc->len, NULL, drpc->cred, drpc->p);
5380 NFSD_DEBUG(4, "start_allocatedsdorpc: err=%d\n", drpc->err);
5384 nfsrv_allocatedsrpc(fhandle_t *fhp, off_t off, off_t len, struct ucred *cred,
5385 NFSPROC_T *p, struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5388 struct nfsrvallocatedsdorpc *drpc, *tdrpc = NULL;
5390 int error, i, ret, timo;
5392 NFSD_DEBUG(4, "in nfsrv_allocatedsrpc\n");
5395 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5399 * Do the allocate RPC for every DS, using a separate kernel process
5400 * for every DS except the last one.
5403 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5405 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5414 if (nfs_pnfsiothreads != 0) {
5415 ret = nfs_pnfsio(start_allocatedsdorpc, tdrpc);
5416 NFSD_DEBUG(4, "nfsrv_allocatedsrpc: nfs_pnfsio=%d\n",
5420 ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, NULL,
5422 if (nfsds_failerr(ret) && *failposp == -1)
5424 else if (error == 0 && ret != 0)
5430 ret = nfsrv_allocatedsdorpc(*nmpp, fhp, off, len, &na, cred, p);
5431 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5432 *failposp = mirrorcnt - 1;
5433 else if (error == 0 && ret != 0)
5436 error = nfsrv_setextattr(vp, &na, p);
5437 NFSD_DEBUG(4, "nfsrv_allocatedsrpc: aft setextat=%d\n", error);
5439 timo = hz / 50; /* Wait for 20msec. */
5442 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5443 /* Wait for RPCs on separate threads to complete. */
5444 while (tdrpc->inprog != 0 && tdrpc->done == 0)
5445 tsleep(&tdrpc->tsk, PVFS, "srvalds", timo);
5446 if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5448 else if (error == 0 && tdrpc->err != 0)
5456 nfsrv_setattrdsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
5457 struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap,
5458 struct nfsvattr *dsnap)
5461 struct nfsrv_descript *nd;
5463 nfsattrbit_t attrbits;
5466 NFSD_DEBUG(4, "in nfsrv_setattrdsdorpc\n");
5467 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5469 * Use a stateid where other is an alternating 01010 pattern and
5470 * seqid is 0xffffffff. This value is not defined as special by
5471 * the RFC and is used by the FreeBSD NFS server to indicate an
5472 * MDS->DS proxy operation.
5474 st.other[0] = 0x55555555;
5475 st.other[1] = 0x55555555;
5476 st.other[2] = 0x55555555;
5477 st.seqid = 0xffffffff;
5478 nfscl_reqstart(nd, NFSPROC_SETATTR, nmp, (u_int8_t *)fhp, sizeof(*fhp),
5479 NULL, NULL, 0, 0, false);
5480 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5481 nfscl_fillsattr(nd, &nap->na_vattr, vp, NFSSATTR_FULL, 0);
5483 /* Do a Getattr for the attributes that change due to writing. */
5484 NFSZERO_ATTRBIT(&attrbits);
5485 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
5486 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
5487 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
5488 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
5489 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
5490 NFSM_BUILD(tl, u_int32_t *, NFSX_UNSIGNED);
5491 *tl = txdr_unsigned(NFSV4OP_GETATTR);
5492 (void) nfsrv_putattrbit(nd, &attrbits);
5493 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
5494 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5499 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattrrpc=%d\n",
5501 /* Get rid of weak cache consistency data for now. */
5502 if ((nd->nd_flag & (ND_NOMOREDATA | ND_NFSV4 | ND_V4WCCATTR)) ==
5503 (ND_NFSV4 | ND_V4WCCATTR)) {
5504 error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL,
5505 NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5506 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: wcc attr=%d\n", error);
5510 * Get rid of Op# and status for next op.
5512 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5514 nd->nd_flag |= ND_NOMOREDATA;
5516 error = nfsrv_getattrbits(nd, &attrbits, NULL, NULL);
5519 if (nd->nd_repstat != 0)
5520 error = nd->nd_repstat;
5522 * Get the Change, Size, Access Time and Modify Time attributes and set
5523 * on the Metadata file, so its attributes will be what the file's
5524 * would be if it had been written.
5527 NFSM_DISSECT(tl, uint32_t *, 2 * NFSX_UNSIGNED);
5528 error = nfsv4_loadattr(nd, NULL, dsnap, NULL, NULL, 0, NULL,
5529 NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL, NULL, NULL);
5531 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc: aft setattr loadattr=%d\n", error);
5533 m_freem(nd->nd_mrep);
5535 NFSD_DEBUG(4, "nfsrv_setattrdsdorpc error=%d\n", error);
5539 struct nfsrvsetattrdsdorpc {
5544 struct nfsmount *nmp;
5549 struct nfsvattr dsna;
5554 * Start up the thread that will execute nfsrv_setattrdsdorpc().
5557 start_setattrdsdorpc(void *arg, int pending)
5559 struct nfsrvsetattrdsdorpc *drpc;
5561 drpc = (struct nfsrvsetattrdsdorpc *)arg;
5562 drpc->err = nfsrv_setattrdsdorpc(&drpc->fh, drpc->cred, drpc->p,
5563 drpc->vp, drpc->nmp, &drpc->na, &drpc->dsna);
5568 nfsrv_setattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
5569 struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt,
5570 struct nfsvattr *nap, int *failposp)
5572 struct nfsrvsetattrdsdorpc *drpc, *tdrpc = NULL;
5574 int error, i, ret, timo;
5576 NFSD_DEBUG(4, "in nfsrv_setattrdsrpc\n");
5579 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5583 * Do the setattr RPC for every DS, using a separate kernel process
5584 * for every DS except the last one.
5587 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5590 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5598 if (nfs_pnfsiothreads != 0) {
5599 ret = nfs_pnfsio(start_setattrdsdorpc, tdrpc);
5600 NFSD_DEBUG(4, "nfsrv_setattrdsrpc: nfs_pnfsio=%d\n",
5604 ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap,
5606 if (nfsds_failerr(ret) && *failposp == -1)
5608 else if (error == 0 && ret != 0)
5614 ret = nfsrv_setattrdsdorpc(fhp, cred, p, vp, *nmpp, nap, &na);
5615 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5616 *failposp = mirrorcnt - 1;
5617 else if (error == 0 && ret != 0)
5620 error = nfsrv_setextattr(vp, &na, p);
5621 NFSD_DEBUG(4, "nfsrv_setattrdsrpc: aft setextat=%d\n", error);
5623 timo = hz / 50; /* Wait for 20msec. */
5626 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5627 /* Wait for RPCs on separate threads to complete. */
5628 while (tdrpc->inprog != 0 && tdrpc->done == 0)
5629 tsleep(&tdrpc->tsk, PVFS, "srvsads", timo);
5630 if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5632 else if (error == 0 && tdrpc->err != 0)
5640 * Do a Setattr of an NFSv4 ACL on the DS file.
5643 nfsrv_setacldsdorpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
5644 struct vnode *vp, struct nfsmount *nmp, struct acl *aclp)
5646 struct nfsrv_descript *nd;
5648 nfsattrbit_t attrbits;
5651 NFSD_DEBUG(4, "in nfsrv_setacldsdorpc\n");
5652 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5654 * Use a stateid where other is an alternating 01010 pattern and
5655 * seqid is 0xffffffff. This value is not defined as special by
5656 * the RFC and is used by the FreeBSD NFS server to indicate an
5657 * MDS->DS proxy operation.
5659 st.other[0] = 0x55555555;
5660 st.other[1] = 0x55555555;
5661 st.other[2] = 0x55555555;
5662 st.seqid = 0xffffffff;
5663 nfscl_reqstart(nd, NFSPROC_SETACL, nmp, (u_int8_t *)fhp, sizeof(*fhp),
5664 NULL, NULL, 0, 0, false);
5665 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5666 NFSZERO_ATTRBIT(&attrbits);
5667 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_ACL);
5669 * The "vp" argument to nfsv4_fillattr() is only used for vnode_type(),
5670 * so passing in the metadata "vp" will be ok, since it is of
5671 * the same type (VREG).
5673 nfsv4_fillattr(nd, NULL, vp, aclp, NULL, NULL, 0, &attrbits, NULL,
5674 NULL, 0, 0, 0, 0, 0, NULL);
5675 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
5676 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5681 NFSD_DEBUG(4, "nfsrv_setacldsdorpc: aft setaclrpc=%d\n",
5683 error = nd->nd_repstat;
5684 m_freem(nd->nd_mrep);
5689 struct nfsrvsetacldsdorpc {
5694 struct nfsmount *nmp;
5703 * Start up the thread that will execute nfsrv_setacldsdorpc().
5706 start_setacldsdorpc(void *arg, int pending)
5708 struct nfsrvsetacldsdorpc *drpc;
5710 drpc = (struct nfsrvsetacldsdorpc *)arg;
5711 drpc->err = nfsrv_setacldsdorpc(&drpc->fh, drpc->cred, drpc->p,
5712 drpc->vp, drpc->nmp, drpc->aclp);
5717 nfsrv_setacldsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
5718 struct vnode *vp, struct nfsmount **nmpp, int mirrorcnt, struct acl *aclp,
5721 struct nfsrvsetacldsdorpc *drpc, *tdrpc = NULL;
5722 int error, i, ret, timo;
5724 NFSD_DEBUG(4, "in nfsrv_setacldsrpc\n");
5727 tdrpc = drpc = malloc(sizeof(*drpc) * (mirrorcnt - 1), M_TEMP,
5731 * Do the setattr RPC for every DS, using a separate kernel process
5732 * for every DS except the last one.
5735 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5738 NFSBCOPY(fhp, &tdrpc->fh, sizeof(*fhp));
5746 if (nfs_pnfsiothreads != 0) {
5747 ret = nfs_pnfsio(start_setacldsdorpc, tdrpc);
5748 NFSD_DEBUG(4, "nfsrv_setacldsrpc: nfs_pnfsio=%d\n",
5752 ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp,
5754 if (nfsds_failerr(ret) && *failposp == -1)
5756 else if (error == 0 && ret != 0)
5762 ret = nfsrv_setacldsdorpc(fhp, cred, p, vp, *nmpp, aclp);
5763 if (nfsds_failerr(ret) && *failposp == -1 && mirrorcnt > 1)
5764 *failposp = mirrorcnt - 1;
5765 else if (error == 0 && ret != 0)
5767 NFSD_DEBUG(4, "nfsrv_setacldsrpc: aft setextat=%d\n", error);
5769 timo = hz / 50; /* Wait for 20msec. */
5772 for (i = 0; i < mirrorcnt - 1; i++, tdrpc++) {
5773 /* Wait for RPCs on separate threads to complete. */
5774 while (tdrpc->inprog != 0 && tdrpc->done == 0)
5775 tsleep(&tdrpc->tsk, PVFS, "srvacds", timo);
5776 if (nfsds_failerr(tdrpc->err) && *failposp == -1)
5778 else if (error == 0 && tdrpc->err != 0)
5786 * Getattr call to the DS for the attributes that change due to writing.
5789 nfsrv_getattrdsrpc(fhandle_t *fhp, struct ucred *cred, NFSPROC_T *p,
5790 struct vnode *vp, struct nfsmount *nmp, struct nfsvattr *nap)
5792 struct nfsrv_descript *nd;
5794 nfsattrbit_t attrbits;
5796 NFSD_DEBUG(4, "in nfsrv_getattrdsrpc\n");
5797 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5798 nfscl_reqstart(nd, NFSPROC_GETATTR, nmp, (u_int8_t *)fhp,
5799 sizeof(fhandle_t), NULL, NULL, 0, 0, false);
5800 NFSZERO_ATTRBIT(&attrbits);
5801 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SIZE);
5802 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_CHANGE);
5803 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEACCESS);
5804 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_TIMEMODIFY);
5805 NFSSETBIT_ATTRBIT(&attrbits, NFSATTRBIT_SPACEUSED);
5806 (void) nfsrv_putattrbit(nd, &attrbits);
5807 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
5808 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5813 NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft getattrrpc=%d\n",
5815 if (nd->nd_repstat == 0) {
5816 error = nfsv4_loadattr(nd, NULL, nap, NULL, NULL, 0,
5817 NULL, NULL, NULL, NULL, NULL, 0, NULL, NULL, NULL,
5820 * We can only save the updated values in the extended
5821 * attribute if the vp is exclusively locked.
5822 * This should happen when any of the following operations
5823 * occur on the vnode:
5824 * Close, Delegreturn, LayoutCommit, LayoutReturn
5825 * As such, the updated extended attribute should get saved
5826 * before nfsrv_checkdsattr() returns 0 and allows the cached
5827 * attributes to be returned without calling this function.
5829 if (error == 0 && VOP_ISLOCKED(vp) == LK_EXCLUSIVE) {
5830 error = nfsrv_setextattr(vp, nap, p);
5831 NFSD_DEBUG(4, "nfsrv_getattrdsrpc: aft setextat=%d\n",
5835 error = nd->nd_repstat;
5836 m_freem(nd->nd_mrep);
5838 NFSD_DEBUG(4, "nfsrv_getattrdsrpc error=%d\n", error);
5843 * Seek call to a DS.
5846 nfsrv_seekdsrpc(fhandle_t *fhp, off_t *offp, int content, bool *eofp,
5847 struct ucred *cred, NFSPROC_T *p, struct nfsmount *nmp)
5850 struct nfsrv_descript *nd;
5854 NFSD_DEBUG(4, "in nfsrv_seekdsrpc\n");
5856 * Use a stateid where other is an alternating 01010 pattern and
5857 * seqid is 0xffffffff. This value is not defined as special by
5858 * the RFC and is used by the FreeBSD NFS server to indicate an
5859 * MDS->DS proxy operation.
5861 st.other[0] = 0x55555555;
5862 st.other[1] = 0x55555555;
5863 st.other[2] = 0x55555555;
5864 st.seqid = 0xffffffff;
5865 nd = malloc(sizeof(*nd), M_TEMP, M_WAITOK | M_ZERO);
5866 nfscl_reqstart(nd, NFSPROC_SEEKDS, nmp, (u_int8_t *)fhp,
5867 sizeof(fhandle_t), NULL, NULL, 0, 0, false);
5868 nfsm_stateidtom(nd, &st, NFSSTATEID_PUTSTATEID);
5869 NFSM_BUILD(tl, uint32_t *, NFSX_HYPER + NFSX_UNSIGNED);
5870 txdr_hyper(*offp, tl); tl += 2;
5871 *tl = txdr_unsigned(content);
5872 error = newnfs_request(nd, nmp, NULL, &nmp->nm_sockreq, NULL, p, cred,
5873 NFS_PROG, NFS_VER4, NULL, 1, NULL, NULL);
5878 NFSD_DEBUG(4, "nfsrv_seekdsrpc: aft seekrpc=%d\n", nd->nd_repstat);
5879 if (nd->nd_repstat == 0) {
5880 NFSM_DISSECT(tl, uint32_t *, NFSX_UNSIGNED + NFSX_HYPER);
5881 if (*tl++ == newnfs_true)
5885 *offp = fxdr_hyper(tl);
5887 error = nd->nd_repstat;
5889 m_freem(nd->nd_mrep);
5891 NFSD_DEBUG(4, "nfsrv_seekdsrpc error=%d\n", error);
5896 * Get the device id and file handle for a DS file.
5899 nfsrv_dsgetdevandfh(struct vnode *vp, NFSPROC_T *p, int *mirrorcntp,
5900 fhandle_t *fhp, char *devid)
5906 buf = malloc(buflen, M_TEMP, M_WAITOK);
5907 error = nfsrv_dsgetsockmnt(vp, 0, buf, &buflen, mirrorcntp, p, NULL,
5908 fhp, devid, NULL, NULL, NULL, NULL, NULL, NULL);
5914 * Do a Lookup against the DS for the filename.
5917 nfsrv_pnfslookupds(struct vnode *vp, struct vnode *dvp, struct pnfsdsfile *pf,
5918 struct vnode **nvpp, NFSPROC_T *p)
5920 struct nameidata named;
5921 struct ucred *tcred;
5927 tcred = newnfs_getcred();
5928 named.ni_cnd.cn_nameiop = LOOKUP;
5929 named.ni_cnd.cn_lkflags = LK_SHARED | LK_RETRY;
5930 named.ni_cnd.cn_cred = tcred;
5931 named.ni_cnd.cn_thread = p;
5932 named.ni_cnd.cn_flags = ISLASTCN | LOCKPARENT | LOCKLEAF | SAVENAME;
5933 nfsvno_setpathbuf(&named, &bufp, &hashp);
5934 named.ni_cnd.cn_nameptr = bufp;
5935 named.ni_cnd.cn_namelen = strlen(pf->dsf_filename);
5936 strlcpy(bufp, pf->dsf_filename, NAME_MAX);
5937 NFSD_DEBUG(4, "nfsrv_pnfslookupds: filename=%s\n", bufp);
5938 error = VOP_LOOKUP(dvp, &nvp, &named.ni_cnd);
5939 NFSD_DEBUG(4, "nfsrv_pnfslookupds: aft LOOKUP=%d\n", error);
5941 nfsvno_relpathbuf(&named);
5944 NFSD_DEBUG(4, "eo nfsrv_pnfslookupds=%d\n", error);
5949 * Set the file handle to the correct one.
5952 nfsrv_pnfssetfh(struct vnode *vp, struct pnfsdsfile *pf, char *devid,
5953 char *fnamep, struct vnode *nvp, NFSPROC_T *p)
5959 NFSBCOPY(np->n_fhp->nfh_fh, &pf->dsf_fh, NFSX_MYFH);
5961 * We can only do a vn_set_extattr() if the vnode is exclusively
5962 * locked and vn_start_write() has been done. If devid != NULL or
5963 * fnamep != NULL or the vnode is shared locked, vn_start_write()
5964 * may not have been done.
5965 * If not done now, it will be done on a future call.
5967 if (devid == NULL && fnamep == NULL && NFSVOPISLOCKED(vp) ==
5969 ret = vn_extattr_set(vp, IO_NODELOCKED,
5970 EXTATTR_NAMESPACE_SYSTEM, "pnfsd.dsfile", sizeof(*pf),
5972 NFSD_DEBUG(4, "eo nfsrv_pnfssetfh=%d\n", ret);
5976 * Cause RPCs waiting on "nmp" to fail. This is called for a DS mount point
5977 * when the DS has failed.
5980 nfsrv_killrpcs(struct nfsmount *nmp)
5984 * Call newnfs_nmcancelreqs() to cause
5985 * any RPCs in progress on the mount point to
5987 * This will cause any process waiting for an
5988 * RPC to complete while holding a vnode lock
5989 * on the mounted-on vnode (such as "df" or
5990 * a non-forced "umount") to fail.
5991 * This will unlock the mounted-on vnode so
5992 * a forced dismount can succeed.
5993 * The NFSMNTP_CANCELRPCS flag should be set when this function is
5996 newnfs_nmcancelreqs(nmp);
6000 * Sum up the statfs info for each of the DSs, so that the client will
6001 * receive the total for all DSs.
6004 nfsrv_pnfsstatfs(struct statfs *sf, struct mount *mp)
6007 struct nfsdevice *ds;
6008 struct vnode **dvpp, **tdvpp, *dvp;
6010 int cnt, error = 0, i;
6012 if (nfsrv_devidcnt <= 0)
6014 dvpp = mallocarray(nfsrv_devidcnt, sizeof(*dvpp), M_TEMP, M_WAITOK);
6015 tsf = malloc(sizeof(*tsf), M_TEMP, M_WAITOK);
6017 /* Get an array of the dvps for the DSs. */
6021 /* First, search for matches for same file system. */
6022 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
6023 if (ds->nfsdev_nmp != NULL && ds->nfsdev_mdsisset != 0 &&
6024 fsidcmp(&ds->nfsdev_mdsfsid, &mp->mnt_stat.f_fsid) == 0) {
6025 if (++i > nfsrv_devidcnt)
6027 *tdvpp++ = ds->nfsdev_dvp;
6031 * If no matches for same file system, total all servers not assigned
6035 TAILQ_FOREACH(ds, &nfsrv_devidhead, nfsdev_list) {
6036 if (ds->nfsdev_nmp != NULL &&
6037 ds->nfsdev_mdsisset == 0) {
6038 if (++i > nfsrv_devidcnt)
6040 *tdvpp++ = ds->nfsdev_dvp;
6047 /* Do a VFS_STATFS() for each of the DSs and sum them up. */
6049 for (i = 0; i < cnt && error == 0; i++) {
6051 error = VFS_STATFS(dvp->v_mount, tsf);
6053 if (sf->f_bsize == 0) {
6054 if (tsf->f_bsize > 0)
6055 sf->f_bsize = tsf->f_bsize;
6059 if (tsf->f_blocks > 0) {
6060 if (sf->f_bsize != tsf->f_bsize) {
6061 tot = tsf->f_blocks * tsf->f_bsize;
6062 sf->f_blocks += (tot / sf->f_bsize);
6064 sf->f_blocks += tsf->f_blocks;
6066 if (tsf->f_bfree > 0) {
6067 if (sf->f_bsize != tsf->f_bsize) {
6068 tot = tsf->f_bfree * tsf->f_bsize;
6069 sf->f_bfree += (tot / sf->f_bsize);
6071 sf->f_bfree += tsf->f_bfree;
6073 if (tsf->f_bavail > 0) {
6074 if (sf->f_bsize != tsf->f_bsize) {
6075 tot = tsf->f_bavail * tsf->f_bsize;
6076 sf->f_bavail += (tot / sf->f_bsize);
6078 sf->f_bavail += tsf->f_bavail;
6091 nfsrv_setacl(struct vnode *vp, NFSACL_T *aclp, struct ucred *cred, NFSPROC_T *p)
6095 if (nfsrv_useacl == 0 || nfs_supportsnfsv4acls(vp) == 0) {
6096 error = NFSERR_ATTRNOTSUPP;
6100 * With NFSv4 ACLs, chmod(2) may need to add additional entries.
6101 * Make sure it has enough room for that - splitting every entry
6102 * into two and appending "canonical six" entries at the end.
6103 * Cribbed out of kern/vfs_acl.c - Rick M.
6105 if (aclp->acl_cnt > (ACL_MAX_ENTRIES - 6) / 2) {
6106 error = NFSERR_ATTRNOTSUPP;
6109 error = VOP_SETACL(vp, ACL_TYPE_NFS4, aclp, cred, p);
6111 error = nfsrv_dssetacl(vp, aclp, cred, p);
6112 if (error == ENOENT)
6122 * Seek vnode op call (actually it is a VOP_IOCTL()).
6123 * This function is called with the vnode locked, but unlocks and vrele()s
6124 * the vp before returning.
6127 nfsvno_seek(struct nfsrv_descript *nd, struct vnode *vp, u_long cmd,
6128 off_t *offp, int content, bool *eofp, struct ucred *cred, NFSPROC_T *p)
6133 ASSERT_VOP_LOCKED(vp, "nfsvno_seek vp");
6135 * Attempt to seek on a DS file. A return of ENOENT implies
6136 * there is no DS file to seek on.
6138 error = nfsrv_proxyds(vp, 0, 0, cred, p, NFSPROC_SEEKDS, NULL,
6139 NULL, NULL, NULL, NULL, offp, content, eofp);
6140 if (error != ENOENT) {
6146 * Do the VOP_IOCTL() call. For the case where *offp == file_size,
6147 * VOP_IOCTL() will return ENXIO. However, the correct reply for
6148 * NFSv4.2 is *eofp == true and error == 0 for this case.
6151 error = VOP_IOCTL(vp, cmd, offp, 0, cred, p);
6153 if (error == ENXIO || (error == 0 && cmd == FIOSEEKHOLE)) {
6154 /* Handle the cases where we might be at EOF. */
6155 ret = nfsvno_getattr(vp, &at, nd, p, 0, NULL);
6156 if (ret == 0 && *offp == at.na_size) {
6160 if (ret != 0 && error == 0)
6169 * Allocate vnode op call.
6172 nfsvno_allocate(struct vnode *vp, off_t off, off_t len, struct ucred *cred,
6177 ASSERT_VOP_ELOCKED(vp, "nfsvno_allocate vp");
6179 * Attempt to allocate on a DS file. A return of ENOENT implies
6180 * there is no DS file to allocate on.
6182 error = nfsrv_proxyds(vp, off, 0, cred, p, NFSPROC_ALLOCATE, NULL,
6183 NULL, NULL, NULL, NULL, &len, 0, NULL);
6184 if (error != ENOENT)
6189 * Do the actual VOP_ALLOCATE(), looping a reasonable number of
6190 * times to achieve completion.
6193 while (error == 0 && len > 0 && trycnt++ < 20)
6194 error = VOP_ALLOCATE(vp, &off, &len);
6195 if (error == 0 && len > 0)
6202 * Get Extended Atribute vnode op into an mbuf list.
6205 nfsvno_getxattr(struct vnode *vp, char *name, uint32_t maxresp,
6206 struct ucred *cred, struct thread *p, struct mbuf **mpp,
6207 struct mbuf **mpendp, int *lenp)
6210 struct uio io, *uiop = &io;
6211 struct mbuf *m, *m2;
6212 int alen, error, len, tlen;
6215 /* First, find out the size of the extended attribute. */
6216 error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL,
6219 return (NFSERR_NOXATTR);
6220 if (siz > maxresp - NFS_MAXXDR)
6221 return (NFSERR_XATTR2BIG);
6223 tlen = NFSM_RNDUP(len);
6225 uiop->uio_iovcnt = nfsrv_createiovec(tlen, &m, &m2, &iv);
6228 uiop->uio_iovcnt = 0;
6229 uiop->uio_iov = iv = NULL;
6232 uiop->uio_offset = 0;
6233 uiop->uio_resid = tlen;
6234 uiop->uio_rw = UIO_READ;
6235 uiop->uio_segflg = UIO_SYSSPACE;
6238 error = mac_vnode_check_getextattr(cred, vp, EXTATTR_NAMESPACE_USER,
6245 error = VOP_GETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop,
6249 if (uiop->uio_resid > 0) {
6251 len = tlen - uiop->uio_resid;
6252 tlen = NFSM_RNDUP(len);
6254 printf("nfsvno_getxattr: weird size read\n");
6258 } else if (alen != tlen || tlen != len)
6259 m2 = nfsrv_adj(m, alen - tlen, tlen - len);
6277 * Set Extended attribute vnode op from an mbuf list.
6280 nfsvno_setxattr(struct vnode *vp, char *name, int len, struct mbuf *m,
6281 char *cp, struct ucred *cred, struct thread *p)
6284 struct uio uio, *uiop = &uio;
6289 error = mac_vnode_check_setextattr(cred, vp, EXTATTR_NAMESPACE_USER,
6295 uiop->uio_rw = UIO_WRITE;
6296 uiop->uio_segflg = UIO_SYSSPACE;
6298 uiop->uio_offset = 0;
6299 uiop->uio_resid = len;
6301 error = nfsrv_createiovecw(len, m, cp, &iv, &cnt);
6303 uiop->uio_iovcnt = cnt;
6305 uiop->uio_iov = iv = NULL;
6306 uiop->uio_iovcnt = 0;
6309 error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, uiop,
6320 * Remove Extended attribute vnode op.
6323 nfsvno_rmxattr(struct nfsrv_descript *nd, struct vnode *vp, char *name,
6324 struct ucred *cred, struct thread *p)
6329 * Get rid of any delegations. I am not sure why this is required,
6330 * but RFC-8276 says so.
6332 error = nfsrv_checkremove(vp, 0, nd, nd->nd_clientid, p);
6336 error = mac_vnode_check_deleteextattr(cred, vp, EXTATTR_NAMESPACE_USER,
6342 error = VOP_DELETEEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, cred, p);
6343 if (error == EOPNOTSUPP)
6344 error = VOP_SETEXTATTR(vp, EXTATTR_NAMESPACE_USER, name, NULL,
6352 * List Extended Atribute vnode op into an mbuf list.
6355 nfsvno_listxattr(struct vnode *vp, uint64_t cookie, struct ucred *cred,
6356 struct thread *p, u_char **bufp, uint32_t *lenp, bool *eofp)
6364 /* First, find out the size of the extended attribute. */
6365 error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, NULL, &siz, cred,
6368 return (NFSERR_NOXATTR);
6369 if (siz <= cookie) {
6374 if (siz > cookie + *lenp) {
6375 siz = cookie + *lenp;
6379 /* Just choose a sanity limit of 10Mbytes for malloc(M_TEMP). */
6380 if (siz > 10 * 1024 * 1024) {
6381 error = NFSERR_XATTR2BIG;
6384 *bufp = malloc(siz, M_TEMP, M_WAITOK);
6385 iv.iov_base = *bufp;
6391 io.uio_rw = UIO_READ;
6392 io.uio_segflg = UIO_SYSSPACE;
6395 error = mac_vnode_check_listextattr(cred, vp, EXTATTR_NAMESPACE_USER);
6400 error = VOP_LISTEXTATTR(vp, EXTATTR_NAMESPACE_USER, &io, NULL, cred,
6404 if (io.uio_resid > 0)
6405 siz -= io.uio_resid;
6410 free(*bufp, M_TEMP);
6417 extern int (*nfsd_call_nfsd)(struct thread *, struct nfssvc_args *);
6420 * Called once to initialize data structures...
6423 nfsd_modevent(module_t mod, int type, void *data)
6426 static int loaded = 0;
6433 for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
6434 mtx_init(&nfsrchash_table[i].mtx, "nfsrtc", NULL,
6436 mtx_init(&nfsrcahash_table[i].mtx, "nfsrtca", NULL,
6439 mtx_init(&nfsrc_udpmtx, "nfsuc", NULL, MTX_DEF);
6440 mtx_init(&nfs_v4root_mutex, "nfs4rt", NULL, MTX_DEF);
6441 mtx_init(&nfsv4root_mnt.mnt_mtx, "nfs4mnt", NULL, MTX_DEF);
6442 mtx_init(&nfsrv_dontlistlock_mtx, "nfs4dnl", NULL, MTX_DEF);
6443 mtx_init(&nfsrv_recalllock_mtx, "nfs4rec", NULL, MTX_DEF);
6444 lockinit(&nfsv4root_mnt.mnt_explock, PVFS, "explock", 0, 0);
6451 #ifdef VV_DISABLEDELEG
6452 vn_deleg_ops.vndeleg_recall = nfsd_recalldelegation;
6453 vn_deleg_ops.vndeleg_disable = nfsd_disabledelegation;
6455 nfsd_call_servertimer = nfsrv_servertimer;
6456 nfsd_call_nfsd = nfssvc_nfsd;
6461 if (newnfs_numnfsd != 0) {
6466 #ifdef VV_DISABLEDELEG
6467 vn_deleg_ops.vndeleg_recall = NULL;
6468 vn_deleg_ops.vndeleg_disable = NULL;
6470 nfsd_call_servertimer = NULL;
6471 nfsd_call_nfsd = NULL;
6473 /* Clean out all NFSv4 state. */
6474 nfsrv_throwawayallstate(curthread);
6476 /* Clean the NFS server reply cache */
6477 nfsrvd_cleancache();
6479 /* Free up the krpc server pool. */
6480 if (nfsrvd_pool != NULL)
6481 svcpool_destroy(nfsrvd_pool);
6483 /* and get rid of the locks */
6484 for (i = 0; i < NFSRVCACHE_HASHSIZE; i++) {
6485 mtx_destroy(&nfsrchash_table[i].mtx);
6486 mtx_destroy(&nfsrcahash_table[i].mtx);
6488 mtx_destroy(&nfsrc_udpmtx);
6489 mtx_destroy(&nfs_v4root_mutex);
6490 mtx_destroy(&nfsv4root_mnt.mnt_mtx);
6491 mtx_destroy(&nfsrv_dontlistlock_mtx);
6492 mtx_destroy(&nfsrv_recalllock_mtx);
6493 for (i = 0; i < nfsrv_sessionhashsize; i++)
6494 mtx_destroy(&nfssessionhash[i].mtx);
6495 if (nfslayouthash != NULL) {
6496 for (i = 0; i < nfsrv_layouthashsize; i++)
6497 mtx_destroy(&nfslayouthash[i].mtx);
6498 free(nfslayouthash, M_NFSDSESSION);
6500 lockdestroy(&nfsv4root_mnt.mnt_explock);
6501 free(nfsclienthash, M_NFSDCLIENT);
6502 free(nfslockhash, M_NFSDLOCKFILE);
6503 free(nfssessionhash, M_NFSDSESSION);
6515 static moduledata_t nfsd_mod = {
6520 DECLARE_MODULE(nfsd, nfsd_mod, SI_SUB_VFS, SI_ORDER_ANY);
6522 /* So that loader and kldload(2) can find us, wherever we are.. */
6523 MODULE_VERSION(nfsd, 1);
6524 MODULE_DEPEND(nfsd, nfscommon, 1, 1, 1);
6525 MODULE_DEPEND(nfsd, nfslockd, 1, 1, 1);
6526 MODULE_DEPEND(nfsd, krpc, 1, 1, 1);
6527 MODULE_DEPEND(nfsd, nfssvc, 1, 1, 1);